Home EDA Trend Clustering Mapping Prediction Plate Number Preprocessing

1. Dataset Preparing (Tidak perlu diubah)

Dalam bagian ini akan menyiapkan dataset sampai siap untuk diolah atau dipreprocessing.
Tensorflow: Package untuk machine learning / AI
Numpy: Package untuk mengolah array
Pandas: Package untuk mengolah dataset
Pickle: Package untuk menyimpan variabel

# Mengambil Package2 yang akan digunakan
import tensorflow as tf
import numpy as np
import pandas as pd
import pickle

# Menggabungkan Netto_kg
df_aggregated = pd.read_csv('metadata_aggregated.csv')  # Membaca file
df_aggregated = df_aggregated.groupby(['tanggal', 'kecamatan']).agg(  # Menggabungkan banyak data dengan menjumlahkan 'netto_kg' untuk setiap 'tanggal' dan 'kecamatan' yang unik
    {
        'netto_kg': 'sum',
    }
).reset_index()
df_aggregated.head()

	tanggal	kecamatan	netto_kg
0	01/01/2020	Magelang Selatan	12820
1	01/01/2020	Magelang Tengah	6750
2	01/01/2020	Magelang Utara	11195
3	01/01/2021	Magelang Selatan	18085
4	01/01/2021	Magelang Tengah	12510

df_aggregated.info()  # menampilkan informasi tentang dataframe

<class 'pandas.core.frame.DataFrame'>
RangeIndex: 5249 entries, 0 to 5248
Data columns (total 3 columns):
 #   Column     Non-Null Count  Dtype 
---  ------     --------------  ----- 
 0   tanggal    5249 non-null   object
 1   kecamatan  5249 non-null   object
 2   netto_kg   5249 non-null   int64 
dtypes: int64(1), object(2)
memory usage: 123.1+ KB

df_aggregated['kecamatan'].value_counts()  # menampilkan jumlah data setiap 'kecamatan'

Magelang Utara      1771
Magelang Selatan    1767
Magelang Tengah     1711
Name: kecamatan, dtype: int64

# Mengimport package untuk mengolah waktu dan normalisasi data
from datetime import datetime, timedelta  # package untuk mengatur waktu/tanggal
from sklearn.preprocessing import MinMaxScaler  # package untuk normalisasi (MinMaxScaler)

Normalisasi adalah proses untuk mengubah rentang data menjadi rentang tertentu, salah satu metode Normalisasi adalah MinMaxScaler, dimana rentang data akan menjadi 0 sampai 1 dengan tujuan untuk menyeimbangkan data dan membuat training model tidak memiliki parameter yang terlalu besar.

def min_max_normalize_dataframe(df, columns):  # Untuk normalisasi dataframe
    # Create a MinMaxScaler object
    scaler = MinMaxScaler()

    # Fit the scaler on the specified columns
    scaler.fit(df[columns])

    # Apply min-max normalization to the specified columns
    df[columns] = scaler.transform(df[columns])

    return df, scaler

def denormalize_dataframe(df, columns, scaler):  # Untuk mengembalikan dataframe ke bentuk semula sebelum normalisasi
    # Inverse transform the specified columns
    df[columns] = scaler.inverse_transform(df[columns])

    return df

def fill_missing_dates(df):  # Untuk mengisi nilai kosong pada dataframe
    df['tanggal'] = pd.to_datetime(df['tanggal'], format='%d/%m/%Y', dayfirst=True)  # Mengubah format kolom 'tanggal' menjadi format tanggal, karena bisa saja kolom tersebut belum pada format tanggal
    df = df.set_index('tanggal')  # Mengubah index dataframe menjadi tanggal, hal ini digunakan karena ada fungsi bernama 'interpolate' pada pandas hanya bisa membaca index, sehingga kolom tanggal perlu diubah menjadi index

    # Generate a date range from the minimum to maximum date in the 'tanggal' column
    min_date = df.index.min()  # Menentukan tanggal terkecil (oldest)
    max_date = df.index.max()  # Menentukan tanggal terbesar (newest)
    date_range = pd.date_range(start=min_date, end=max_date)  # Menentukan rentang tanggal dari terkecil hingga terbesar

    df = df.reindex(date_range)  # Menambahkan data berisikan Null (data kosong) ke dataframe yang tidak memuat tanggal pada data_range (tanggal yang kosong)

    df['netto_kg'] = df['netto_kg'].interpolate(method='linear')  # Mengisi nilai yang Null tadi dengan interpolate menggunakan metode 'linear', artinya data diisi dengan rata-rata antara data sebelumnya dan data berikutnya yang diketahui

    df = df.reset_index()  # Mengembalikan tanggal yang tadinya sebagai index kembali menjadi kolom, namun nama kolomnya akan menjadi 'index' bukan 'tanggal'
    df = df.rename(columns={'index': 'tanggal'})  # Mengubah nama kolom 'index' kembali menjadi 'tanggal

    return df

def check_missing_dates(df):  # Untuk menentukan/mengecek apakah ada tanggal kosong pada dataframe, hal ini digunakan sekali saja untuk mengecek
    # Generate a date range from the minimum to maximum date in the 'tanggal' column
    min_date = df['tanggal'].min()  # Menentukan tanggal terkecil
    max_date = df['tanggal'].max()  # Menentukan tanggal terbesar

    date_range = [min_date + timedelta(days=x) for x in range((max_date - min_date).days + 1)]  # Rentang tanggal

    temp_tanggal = df['tanggal'].values
    # Find missing dates
    missing_dates = [date for date in date_range if date not in temp_tanggal]

    # Print missing dates
    if missing_dates:
        print("Missing dates:")
        for date in missing_dates:
            print(date.strftime('%d/%m/%Y'))
    else:
        print("No missing dates found.")

# Membagi dataset ke magelang utara, tengah, dan selatan, hal ini dikarenakan model yang akan kita buat harus terpisah antar kolom, sedangkan semua kecamatan masih berada pada satu kolom yaitu kolom 'kecamatan'
df_utara = df_aggregated[df_aggregated['kecamatan'] == 'Magelang Utara'].drop(columns='kecamatan').reset_index(drop=True)  # memisah dataframe untuk Magelang Utara
df_tengah = df_aggregated[df_aggregated['kecamatan'] == 'Magelang Tengah'].drop(columns='kecamatan').reset_index(drop=True)  # memisah dataframe untuk Magelang Tengah
df_selatan = df_aggregated[df_aggregated['kecamatan'] == 'Magelang Selatan'].drop(columns='kecamatan').reset_index(drop=True)  # Memisah dataframe untuk Magelang Selatan

df_utara = fill_missing_dates(df_utara).rename(columns={'netto_kg': 'magelang_utara'})  # mengisi tanggal kosong untuk Magelang Utara
df_tengah = fill_missing_dates(df_tengah).rename(columns={'netto_kg': 'magelang_tengah'})  # mengisi tanggal kosong untuk Magelang Tengah
df_selatan = fill_missing_dates(df_selatan).rename(columns={'netto_kg': 'magelang_selatan'})  # mengisi tanggal kosong untuk Magelang Selatan

df_timeseries = pd.merge(df_utara, df_tengah, 'left', 'tanggal')  # Menggabungkan kembali ketiga dataframe menjadi setiap kolom
df_timeseries = pd.merge(df_timeseries, df_selatan, 'left', 'tanggal')

df_timeseries, scaler = min_max_normalize_dataframe(df_timeseries, ['magelang_utara', 'magelang_tengah', 'magelang_selatan'])  # Normalisasi data dengan MinMaxScaler
df_timeseries.info()

<class 'pandas.core.frame.DataFrame'>
Int64Index: 1824 entries, 0 to 1823
Data columns (total 4 columns):
 #   Column            Non-Null Count  Dtype         
---  ------            --------------  -----         
 0   tanggal           1824 non-null   datetime64[ns]
 1   magelang_utara    1824 non-null   float64       
 2   magelang_tengah   1824 non-null   float64       
 3   magelang_selatan  1824 non-null   float64       
dtypes: datetime64[ns](1), float64(3)
memory usage: 71.2 KB

df_timeseries.head()  # Menampilkan 5 data pertama

	tanggal	magelang_utara	magelang_tengah	magelang_selatan
0	2019-01-03	0.174009	0.001101	0.116399
1	2019-01-04	0.637558	0.231447	0.123363
2	2019-01-05	0.552547	0.194891	0.173437
3	2019-01-06	0.592419	0.181238	0.283203
4	2019-01-07	0.848929	0.664832	0.528934

df_timeseries.describe()  # Menampilkan informasi statistik

	magelang_utara	magelang_tengah	magelang_selatan
count	1824.000000	1824.000000	1824.000000
mean	0.460543	0.451285	0.516479
std	0.180072	0.169291	0.184549
min	0.000000	0.000000	0.000000
25%	0.363032	0.345959	0.406400
50%	0.479326	0.460802	0.534240
75%	0.584174	0.563312	0.647198
max	1.000000	1.000000	1.000000

# membuat fungsi untuk membuat dataset, dataset harus berisikan input dan output yang tetap semisal input 5 hari sebelumnya untuk memprediksi output 30 hari setelahnya
def make_dataset(df, columns, input_timesteps=10, output_timesteps=10):
    # Iterate through the DataFrame to create input-output pairs
    X = []
    y = []
    for i in range(len(df) - input_timesteps - output_timesteps + 1):  # iterasi untuk setiap data
        X.append(df[columns].iloc[i:i+input_timesteps].values)
        y.append(df[columns].iloc[i+input_timesteps:i+input_timesteps+output_timesteps].values)

    # Convert the input-output pairs into numpy arrays
    X = np.array(X)
    y = np.array(y)

    print("Input shape:", X.shape)
    print("Output shape:", y.shape)

    return X, y

# memperlihatkan jumlah dataset
X_30, y_30 = make_dataset(df_timeseries, ['magelang_utara', 'magelang_tengah', 'magelang_selatan'], input_timesteps=30, output_timesteps=366)  # Data dengan input 30 hari, output 366 hari
X_90, y_90 = make_dataset(df_timeseries, ['magelang_utara', 'magelang_tengah', 'magelang_selatan'], input_timesteps=90, output_timesteps=366)  # Data dengan input 90 hari, output 366 hari
X_366, y_366 = make_dataset(df_timeseries, ['magelang_utara', 'magelang_tengah', 'magelang_selatan'], input_timesteps=366, output_timesteps=366)  # Data dengan input 366 hari, output 366 hari

Input shape: (1429, 30, 3)
Output shape: (1429, 366, 3)
Input shape: (1369, 90, 3)
Output shape: (1369, 366, 3)
Input shape: (1093, 366, 3)
Output shape: (1093, 366, 3)

2. Modelling (Diubah sesuai model)

Setiap model dibagi menjadi 3, pertama inputnya timesteps 30, kedua 90, ketiga 366

a. 30 Days LSTM

Model ini menggunakan 30 hari data sampah untuk memprediksi sampah tahun 2024

from keras.callbacks import ModelCheckpoint

# Membuat Model
model_LSTM = tf.keras.Sequential()  # Model lapisan, model ini akan berisi lapisan-lapisan (layers) machine learning
model_LSTM.add(tf.keras.layers.LSTM(64, return_sequences=True, input_shape=(30, 3)))  # Lapisan LSTM pertama
model_LSTM.add(tf.keras.layers.LSTM(64, return_sequences=True))  # Lapisan LSTM kedua
model_LSTM.add(tf.keras.layers.LSTM(256, return_sequences=False))  # Lapisan LSTM ketiga
model_LSTM.add(tf.keras.layers.Dense(1098, activation='relu'))  # Lapisan Dense untuk output
model_LSTM.add(tf.keras.layers.Reshape((366, 3)))  # Pembentukan output menjadi (366, 3) karena ada 366 hari dan 3 kecamatan
model_LSTM.summary()  # Melihat summary dari model

Model: "sequential_20"
_________________________________________________________________
 Layer (type)                Output Shape              Param #   
=================================================================
 lstm_40 (LSTM)              (None, 30, 64)            17408     
                                                                 
 lstm_41 (LSTM)              (None, 30, 64)            33024     
                                                                 
 lstm_42 (LSTM)              (None, 256)               328704    
                                                                 
 dense_15 (Dense)            (None, 1098)              282186    
                                                                 
 reshape_12 (Reshape)        (None, 366, 3)            0         
                                                                 
=================================================================
Total params: 661322 (2.52 MB)
Trainable params: 661322 (2.52 MB)
Non-trainable params: 0 (0.00 Byte)
_________________________________________________________________

model_LSTM.compile(optimizer='adam', loss='mse')  # Compile model dengan optimizer 'adam' dan loss 'mse'

Secara teknikal, Optimizer adalah metode untuk mengubah bobot setiap parameter dengan 'loss' dan parameter tambahan yang diketahui
ADAM (Adaptive Moment Estimation) adalah salah satu optimizer yang paling sering dipakai karena mudah untuk konvergen ke solusi

checkpoint = ModelCheckpoint('best_LSTM_model.h5', monitor='val_loss', save_best_only=True, verbose=1)  # Checkpoint berfungsi untuk menyimpan model dengan kualitas terbaik

Setiap iterasi, model dengan validation loss terkecil akan disimpan.
Data sendiri akan dibagi menjadi 20% Data Test dan 80% Data Train dengan 'validation_split=0.2'

# Melatih Model
history = model_LSTM.fit(X_30, y_30, epochs=100, validation_split=0.2, callbacks=[checkpoint])

Epoch 1/100
36/36 [==============================] - ETA: 0s - loss: 0.1316
Epoch 1: val_loss improved from inf to 0.09574, saving model to best_LSTM_model.h5
36/36 [==============================] - 14s 215ms/step - loss: 0.1316 - val_loss: 0.0957
Epoch 2/100
36/36 [==============================] - ETA: 0s - loss: 0.1081
Epoch 2: val_loss improved from 0.09574 to 0.09466, saving model to best_LSTM_model.h5
36/36 [==============================] - 5s 147ms/step - loss: 0.1081 - val_loss: 0.0947
Epoch 3/100
36/36 [==============================] - ETA: 0s - loss: 0.1078
Epoch 3: val_loss improved from 0.09466 to 0.09441, saving model to best_LSTM_model.h5
36/36 [==============================] - 6s 177ms/step - loss: 0.1078 - val_loss: 0.0944
Epoch 4/100
36/36 [==============================] - ETA: 0s - loss: 0.1077
Epoch 4: val_loss did not improve from 0.09441
36/36 [==============================] - 5s 146ms/step - loss: 0.1077 - val_loss: 0.0949
Epoch 5/100
36/36 [==============================] - ETA: 0s - loss: 0.1077
Epoch 5: val_loss did not improve from 0.09441
36/36 [==============================] - 6s 178ms/step - loss: 0.1077 - val_loss: 0.0950
Epoch 6/100
36/36 [==============================] - ETA: 0s - loss: 0.1076
Epoch 6: val_loss improved from 0.09441 to 0.09434, saving model to best_LSTM_model.h5
36/36 [==============================] - 5s 147ms/step - loss: 0.1076 - val_loss: 0.0943
Epoch 7/100
36/36 [==============================] - ETA: 0s - loss: 0.1072
Epoch 7: val_loss improved from 0.09434 to 0.09416, saving model to best_LSTM_model.h5
36/36 [==============================] - 7s 187ms/step - loss: 0.1072 - val_loss: 0.0942
Epoch 8/100
36/36 [==============================] - ETA: 0s - loss: 0.1071
Epoch 8: val_loss improved from 0.09416 to 0.09414, saving model to best_LSTM_model.h5
36/36 [==============================] - 5s 148ms/step - loss: 0.1071 - val_loss: 0.0941
Epoch 9/100
36/36 [==============================] - ETA: 0s - loss: 0.1070
Epoch 9: val_loss did not improve from 0.09414
36/36 [==============================] - 6s 172ms/step - loss: 0.1070 - val_loss: 0.0942
Epoch 10/100
36/36 [==============================] - ETA: 0s - loss: 0.1069
Epoch 10: val_loss improved from 0.09414 to 0.09392, saving model to best_LSTM_model.h5
36/36 [==============================] - 6s 158ms/step - loss: 0.1069 - val_loss: 0.0939
Epoch 11/100
36/36 [==============================] - ETA: 0s - loss: 0.1067
Epoch 11: val_loss improved from 0.09392 to 0.09378, saving model to best_LSTM_model.h5
36/36 [==============================] - 6s 157ms/step - loss: 0.1067 - val_loss: 0.0938
Epoch 12/100
36/36 [==============================] - ETA: 0s - loss: 0.1063
Epoch 12: val_loss improved from 0.09378 to 0.09306, saving model to best_LSTM_model.h5
36/36 [==============================] - 6s 171ms/step - loss: 0.1063 - val_loss: 0.0931
Epoch 13/100
36/36 [==============================] - ETA: 0s - loss: 0.1060
Epoch 13: val_loss improved from 0.09306 to 0.09285, saving model to best_LSTM_model.h5
36/36 [==============================] - 5s 147ms/step - loss: 0.1060 - val_loss: 0.0928
Epoch 14/100
36/36 [==============================] - ETA: 0s - loss: 0.1058
Epoch 14: val_loss improved from 0.09285 to 0.09274, saving model to best_LSTM_model.h5
36/36 [==============================] - 7s 210ms/step - loss: 0.1058 - val_loss: 0.0927
Epoch 15/100
36/36 [==============================] - ETA: 0s - loss: 0.1049
Epoch 15: val_loss improved from 0.09274 to 0.09178, saving model to best_LSTM_model.h5
36/36 [==============================] - 5s 148ms/step - loss: 0.1049 - val_loss: 0.0918
Epoch 16/100
36/36 [==============================] - ETA: 0s - loss: 0.1040
Epoch 16: val_loss improved from 0.09178 to 0.09054, saving model to best_LSTM_model.h5
36/36 [==============================] - 6s 180ms/step - loss: 0.1040 - val_loss: 0.0905
Epoch 17/100
36/36 [==============================] - ETA: 0s - loss: 0.1032
Epoch 17: val_loss did not improve from 0.09054
36/36 [==============================] - 5s 147ms/step - loss: 0.1032 - val_loss: 0.0907
Epoch 18/100
36/36 [==============================] - ETA: 0s - loss: 0.1026
Epoch 18: val_loss improved from 0.09054 to 0.08995, saving model to best_LSTM_model.h5
36/36 [==============================] - 6s 180ms/step - loss: 0.1026 - val_loss: 0.0900
Epoch 19/100
36/36 [==============================] - ETA: 0s - loss: 0.1022
Epoch 19: val_loss improved from 0.08995 to 0.08958, saving model to best_LSTM_model.h5
36/36 [==============================] - 5s 148ms/step - loss: 0.1022 - val_loss: 0.0896
Epoch 20/100
36/36 [==============================] - ETA: 0s - loss: 0.1019
Epoch 20: val_loss improved from 0.08958 to 0.08906, saving model to best_LSTM_model.h5
36/36 [==============================] - 8s 215ms/step - loss: 0.1019 - val_loss: 0.0891
Epoch 21/100
36/36 [==============================] - ETA: 0s - loss: 0.1011
Epoch 21: val_loss improved from 0.08906 to 0.08792, saving model to best_LSTM_model.h5
36/36 [==============================] - 5s 149ms/step - loss: 0.1011 - val_loss: 0.0879
Epoch 22/100
36/36 [==============================] - ETA: 0s - loss: 0.0990
Epoch 22: val_loss improved from 0.08792 to 0.08448, saving model to best_LSTM_model.h5
36/36 [==============================] - 6s 163ms/step - loss: 0.0990 - val_loss: 0.0845
Epoch 23/100
36/36 [==============================] - ETA: 0s - loss: 0.0956
Epoch 23: val_loss improved from 0.08448 to 0.08311, saving model to best_LSTM_model.h5
36/36 [==============================] - 6s 161ms/step - loss: 0.0956 - val_loss: 0.0831
Epoch 24/100
36/36 [==============================] - ETA: 0s - loss: 0.0937
Epoch 24: val_loss improved from 0.08311 to 0.08114, saving model to best_LSTM_model.h5
36/36 [==============================] - 5s 150ms/step - loss: 0.0937 - val_loss: 0.0811
Epoch 25/100
36/36 [==============================] - ETA: 0s - loss: 0.0918
Epoch 25: val_loss improved from 0.08114 to 0.07963, saving model to best_LSTM_model.h5
36/36 [==============================] - 6s 179ms/step - loss: 0.0918 - val_loss: 0.0796
Epoch 26/100
36/36 [==============================] - ETA: 0s - loss: 0.0905
Epoch 26: val_loss improved from 0.07963 to 0.07834, saving model to best_LSTM_model.h5
36/36 [==============================] - 5s 148ms/step - loss: 0.0905 - val_loss: 0.0783
Epoch 27/100
36/36 [==============================] - ETA: 0s - loss: 0.0889
Epoch 27: val_loss improved from 0.07834 to 0.07652, saving model to best_LSTM_model.h5
36/36 [==============================] - 6s 179ms/step - loss: 0.0889 - val_loss: 0.0765
Epoch 28/100
36/36 [==============================] - ETA: 0s - loss: 0.0874
Epoch 28: val_loss did not improve from 0.07652
36/36 [==============================] - 5s 146ms/step - loss: 0.0874 - val_loss: 0.0766
Epoch 29/100
36/36 [==============================] - ETA: 0s - loss: 0.0867
Epoch 29: val_loss improved from 0.07652 to 0.07590, saving model to best_LSTM_model.h5
36/36 [==============================] - 7s 188ms/step - loss: 0.0867 - val_loss: 0.0759
Epoch 30/100
36/36 [==============================] - ETA: 0s - loss: 0.0862
Epoch 30: val_loss improved from 0.07590 to 0.07481, saving model to best_LSTM_model.h5
36/36 [==============================] - 5s 149ms/step - loss: 0.0862 - val_loss: 0.0748
Epoch 31/100
36/36 [==============================] - ETA: 0s - loss: 0.0852
Epoch 31: val_loss improved from 0.07481 to 0.07444, saving model to best_LSTM_model.h5
36/36 [==============================] - 6s 182ms/step - loss: 0.0852 - val_loss: 0.0744
Epoch 32/100
36/36 [==============================] - ETA: 0s - loss: 0.0844
Epoch 32: val_loss improved from 0.07444 to 0.07364, saving model to best_LSTM_model.h5
36/36 [==============================] - 5s 149ms/step - loss: 0.0844 - val_loss: 0.0736
Epoch 33/100
36/36 [==============================] - ETA: 0s - loss: 0.0838
Epoch 33: val_loss improved from 0.07364 to 0.07354, saving model to best_LSTM_model.h5
36/36 [==============================] - 6s 180ms/step - loss: 0.0838 - val_loss: 0.0735
Epoch 34/100
36/36 [==============================] - ETA: 0s - loss: 0.0831
Epoch 34: val_loss improved from 0.07354 to 0.07350, saving model to best_LSTM_model.h5
36/36 [==============================] - 5s 147ms/step - loss: 0.0831 - val_loss: 0.0735
Epoch 35/100
36/36 [==============================] - ETA: 0s - loss: 0.0828
Epoch 35: val_loss improved from 0.07350 to 0.07253, saving model to best_LSTM_model.h5
36/36 [==============================] - 6s 180ms/step - loss: 0.0828 - val_loss: 0.0725
Epoch 36/100
36/36 [==============================] - ETA: 0s - loss: 0.0815
Epoch 36: val_loss improved from 0.07253 to 0.07150, saving model to best_LSTM_model.h5
36/36 [==============================] - 5s 148ms/step - loss: 0.0815 - val_loss: 0.0715
Epoch 37/100
36/36 [==============================] - ETA: 0s - loss: 0.0795
Epoch 37: val_loss improved from 0.07150 to 0.06985, saving model to best_LSTM_model.h5
36/36 [==============================] - 6s 155ms/step - loss: 0.0795 - val_loss: 0.0699
Epoch 38/100
36/36 [==============================] - ETA: 0s - loss: 0.0785
Epoch 38: val_loss improved from 0.06985 to 0.06912, saving model to best_LSTM_model.h5
36/36 [==============================] - 6s 169ms/step - loss: 0.0785 - val_loss: 0.0691
Epoch 39/100
36/36 [==============================] - ETA: 0s - loss: 0.0773
Epoch 39: val_loss improved from 0.06912 to 0.06748, saving model to best_LSTM_model.h5
36/36 [==============================] - 5s 148ms/step - loss: 0.0773 - val_loss: 0.0675
Epoch 40/100
36/36 [==============================] - ETA: 0s - loss: 0.0759
Epoch 40: val_loss improved from 0.06748 to 0.06679, saving model to best_LSTM_model.h5
36/36 [==============================] - 6s 179ms/step - loss: 0.0759 - val_loss: 0.0668
Epoch 41/100
36/36 [==============================] - ETA: 0s - loss: 0.0748
Epoch 41: val_loss improved from 0.06679 to 0.06652, saving model to best_LSTM_model.h5
36/36 [==============================] - 5s 148ms/step - loss: 0.0748 - val_loss: 0.0665
Epoch 42/100
36/36 [==============================] - ETA: 0s - loss: 0.0734
Epoch 42: val_loss improved from 0.06652 to 0.06393, saving model to best_LSTM_model.h5
36/36 [==============================] - 6s 178ms/step - loss: 0.0734 - val_loss: 0.0639
Epoch 43/100
36/36 [==============================] - ETA: 0s - loss: 0.0708
Epoch 43: val_loss improved from 0.06393 to 0.06151, saving model to best_LSTM_model.h5
36/36 [==============================] - 5s 147ms/step - loss: 0.0708 - val_loss: 0.0615
Epoch 44/100
36/36 [==============================] - ETA: 0s - loss: 0.0686
Epoch 44: val_loss improved from 0.06151 to 0.06133, saving model to best_LSTM_model.h5
36/36 [==============================] - 6s 177ms/step - loss: 0.0686 - val_loss: 0.0613
Epoch 45/100
36/36 [==============================] - ETA: 0s - loss: 0.0675
Epoch 45: val_loss improved from 0.06133 to 0.05925, saving model to best_LSTM_model.h5
36/36 [==============================] - 5s 149ms/step - loss: 0.0675 - val_loss: 0.0592
Epoch 46/100
36/36 [==============================] - ETA: 0s - loss: 0.0657
Epoch 46: val_loss improved from 0.05925 to 0.05721, saving model to best_LSTM_model.h5
36/36 [==============================] - 6s 180ms/step - loss: 0.0657 - val_loss: 0.0572
Epoch 47/100
36/36 [==============================] - ETA: 0s - loss: 0.0640
Epoch 47: val_loss did not improve from 0.05721
36/36 [==============================] - 5s 149ms/step - loss: 0.0640 - val_loss: 0.0573
Epoch 48/100
36/36 [==============================] - ETA: 0s - loss: 0.0625
Epoch 48: val_loss improved from 0.05721 to 0.05469, saving model to best_LSTM_model.h5
36/36 [==============================] - 6s 159ms/step - loss: 0.0625 - val_loss: 0.0547
Epoch 49/100
36/36 [==============================] - ETA: 0s - loss: 0.0605
Epoch 49: val_loss improved from 0.05469 to 0.05362, saving model to best_LSTM_model.h5
36/36 [==============================] - 6s 166ms/step - loss: 0.0605 - val_loss: 0.0536
Epoch 50/100
36/36 [==============================] - ETA: 0s - loss: 0.0593
Epoch 50: val_loss improved from 0.05362 to 0.05301, saving model to best_LSTM_model.h5
36/36 [==============================] - 5s 147ms/step - loss: 0.0593 - val_loss: 0.0530
Epoch 51/100
36/36 [==============================] - ETA: 0s - loss: 0.0588
Epoch 51: val_loss improved from 0.05301 to 0.05243, saving model to best_LSTM_model.h5
36/36 [==============================] - 6s 179ms/step - loss: 0.0588 - val_loss: 0.0524
Epoch 52/100
36/36 [==============================] - ETA: 0s - loss: 0.0577
Epoch 52: val_loss improved from 0.05243 to 0.05204, saving model to best_LSTM_model.h5
36/36 [==============================] - 5s 148ms/step - loss: 0.0577 - val_loss: 0.0520
Epoch 53/100
36/36 [==============================] - ETA: 0s - loss: 0.0565
Epoch 53: val_loss improved from 0.05204 to 0.05096, saving model to best_LSTM_model.h5
36/36 [==============================] - 8s 219ms/step - loss: 0.0565 - val_loss: 0.0510
Epoch 54/100
36/36 [==============================] - ETA: 0s - loss: 0.0540
Epoch 54: val_loss improved from 0.05096 to 0.04828, saving model to best_LSTM_model.h5
36/36 [==============================] - 5s 147ms/step - loss: 0.0540 - val_loss: 0.0483
Epoch 55/100
36/36 [==============================] - ETA: 0s - loss: 0.0517
Epoch 55: val_loss improved from 0.04828 to 0.04535, saving model to best_LSTM_model.h5
36/36 [==============================] - 6s 180ms/step - loss: 0.0517 - val_loss: 0.0454
Epoch 56/100
36/36 [==============================] - ETA: 0s - loss: 0.0490
Epoch 56: val_loss improved from 0.04535 to 0.04435, saving model to best_LSTM_model.h5
36/36 [==============================] - 5s 145ms/step - loss: 0.0490 - val_loss: 0.0443
Epoch 57/100
36/36 [==============================] - ETA: 0s - loss: 0.0474
Epoch 57: val_loss improved from 0.04435 to 0.04328, saving model to best_LSTM_model.h5
36/36 [==============================] - 6s 168ms/step - loss: 0.0474 - val_loss: 0.0433
Epoch 58/100
36/36 [==============================] - ETA: 0s - loss: 0.0461
Epoch 58: val_loss improved from 0.04328 to 0.04238, saving model to best_LSTM_model.h5
36/36 [==============================] - 6s 155ms/step - loss: 0.0461 - val_loss: 0.0424
Epoch 59/100
36/36 [==============================] - ETA: 0s - loss: 0.0448
Epoch 59: val_loss improved from 0.04238 to 0.04098, saving model to best_LSTM_model.h5
36/36 [==============================] - 6s 157ms/step - loss: 0.0448 - val_loss: 0.0410
Epoch 60/100
36/36 [==============================] - ETA: 0s - loss: 0.0435
Epoch 60: val_loss improved from 0.04098 to 0.04060, saving model to best_LSTM_model.h5
36/36 [==============================] - 6s 170ms/step - loss: 0.0435 - val_loss: 0.0406
Epoch 61/100
36/36 [==============================] - ETA: 0s - loss: 0.0422
Epoch 61: val_loss improved from 0.04060 to 0.03868, saving model to best_LSTM_model.h5
36/36 [==============================] - 5s 148ms/step - loss: 0.0422 - val_loss: 0.0387
Epoch 62/100
36/36 [==============================] - ETA: 0s - loss: 0.0409
Epoch 62: val_loss improved from 0.03868 to 0.03844, saving model to best_LSTM_model.h5
36/36 [==============================] - 6s 180ms/step - loss: 0.0409 - val_loss: 0.0384
Epoch 63/100
36/36 [==============================] - ETA: 0s - loss: 0.0403
Epoch 63: val_loss improved from 0.03844 to 0.03779, saving model to best_LSTM_model.h5
36/36 [==============================] - 5s 146ms/step - loss: 0.0403 - val_loss: 0.0378
Epoch 64/100
36/36 [==============================] - ETA: 0s - loss: 0.0396
Epoch 64: val_loss improved from 0.03779 to 0.03669, saving model to best_LSTM_model.h5
36/36 [==============================] - 6s 178ms/step - loss: 0.0396 - val_loss: 0.0367
Epoch 65/100
36/36 [==============================] - ETA: 0s - loss: 0.0385
Epoch 65: val_loss improved from 0.03669 to 0.03662, saving model to best_LSTM_model.h5
36/36 [==============================] - 5s 148ms/step - loss: 0.0385 - val_loss: 0.0366
Epoch 66/100
36/36 [==============================] - ETA: 0s - loss: 0.0374
Epoch 66: val_loss improved from 0.03662 to 0.03543, saving model to best_LSTM_model.h5
36/36 [==============================] - 6s 179ms/step - loss: 0.0374 - val_loss: 0.0354
Epoch 67/100
36/36 [==============================] - ETA: 0s - loss: 0.0364
Epoch 67: val_loss improved from 0.03543 to 0.03413, saving model to best_LSTM_model.h5
36/36 [==============================] - 5s 149ms/step - loss: 0.0364 - val_loss: 0.0341
Epoch 68/100
36/36 [==============================] - ETA: 0s - loss: 0.0348
Epoch 68: val_loss improved from 0.03413 to 0.03387, saving model to best_LSTM_model.h5
36/36 [==============================] - 6s 181ms/step - loss: 0.0348 - val_loss: 0.0339
Epoch 69/100
36/36 [==============================] - ETA: 0s - loss: 0.0340
Epoch 69: val_loss improved from 0.03387 to 0.03366, saving model to best_LSTM_model.h5
36/36 [==============================] - 5s 149ms/step - loss: 0.0340 - val_loss: 0.0337
Epoch 70/100
36/36 [==============================] - ETA: 0s - loss: 0.0335
Epoch 70: val_loss improved from 0.03366 to 0.03346, saving model to best_LSTM_model.h5
36/36 [==============================] - 6s 179ms/step - loss: 0.0335 - val_loss: 0.0335
Epoch 71/100
36/36 [==============================] - ETA: 0s - loss: 0.0335
Epoch 71: val_loss improved from 0.03346 to 0.03346, saving model to best_LSTM_model.h5
36/36 [==============================] - 5s 146ms/step - loss: 0.0335 - val_loss: 0.0335
Epoch 72/100
36/36 [==============================] - ETA: 0s - loss: 0.0331
Epoch 72: val_loss did not improve from 0.03346
36/36 [==============================] - 5s 151ms/step - loss: 0.0331 - val_loss: 0.0336
Epoch 73/100
36/36 [==============================] - ETA: 0s - loss: 0.0321
Epoch 73: val_loss improved from 0.03346 to 0.03322, saving model to best_LSTM_model.h5
36/36 [==============================] - 6s 170ms/step - loss: 0.0321 - val_loss: 0.0332
Epoch 74/100
36/36 [==============================] - ETA: 0s - loss: 0.0315
Epoch 74: val_loss improved from 0.03322 to 0.03227, saving model to best_LSTM_model.h5
36/36 [==============================] - 5s 149ms/step - loss: 0.0315 - val_loss: 0.0323
Epoch 75/100
36/36 [==============================] - ETA: 0s - loss: 0.0309
Epoch 75: val_loss improved from 0.03227 to 0.03182, saving model to best_LSTM_model.h5
36/36 [==============================] - 6s 178ms/step - loss: 0.0309 - val_loss: 0.0318
Epoch 76/100
36/36 [==============================] - ETA: 0s - loss: 0.0299
Epoch 76: val_loss improved from 0.03182 to 0.03178, saving model to best_LSTM_model.h5
36/36 [==============================] - 5s 149ms/step - loss: 0.0299 - val_loss: 0.0318
Epoch 77/100
36/36 [==============================] - ETA: 0s - loss: 0.0296
Epoch 77: val_loss improved from 0.03178 to 0.03160, saving model to best_LSTM_model.h5
36/36 [==============================] - 6s 178ms/step - loss: 0.0296 - val_loss: 0.0316
Epoch 78/100
36/36 [==============================] - ETA: 0s - loss: 0.0293
Epoch 78: val_loss improved from 0.03160 to 0.03123, saving model to best_LSTM_model.h5
36/36 [==============================] - 5s 149ms/step - loss: 0.0293 - val_loss: 0.0312
Epoch 79/100
36/36 [==============================] - ETA: 0s - loss: 0.0288
Epoch 79: val_loss did not improve from 0.03123
36/36 [==============================] - 7s 184ms/step - loss: 0.0288 - val_loss: 0.0320
Epoch 80/100
36/36 [==============================] - ETA: 0s - loss: 0.0287
Epoch 80: val_loss did not improve from 0.03123
36/36 [==============================] - 5s 146ms/step - loss: 0.0287 - val_loss: 0.0318
Epoch 81/100
36/36 [==============================] - ETA: 0s - loss: 0.0286
Epoch 81: val_loss did not improve from 0.03123
36/36 [==============================] - 6s 173ms/step - loss: 0.0286 - val_loss: 0.0319
Epoch 82/100
36/36 [==============================] - ETA: 0s - loss: 0.0284
Epoch 82: val_loss did not improve from 0.03123
36/36 [==============================] - 5s 147ms/step - loss: 0.0284 - val_loss: 0.0321
Epoch 83/100
36/36 [==============================] - ETA: 0s - loss: 0.0281
Epoch 83: val_loss did not improve from 0.03123
36/36 [==============================] - 6s 157ms/step - loss: 0.0281 - val_loss: 0.0318
Epoch 84/100
36/36 [==============================] - ETA: 0s - loss: 0.0276
Epoch 84: val_loss did not improve from 0.03123
36/36 [==============================] - 6s 168ms/step - loss: 0.0276 - val_loss: 0.0317
Epoch 85/100
36/36 [==============================] - ETA: 0s - loss: 0.0273
Epoch 85: val_loss did not improve from 0.03123
36/36 [==============================] - 5s 148ms/step - loss: 0.0273 - val_loss: 0.0323
Epoch 86/100
36/36 [==============================] - ETA: 0s - loss: 0.0271
Epoch 86: val_loss did not improve from 0.03123
36/36 [==============================] - 8s 212ms/step - loss: 0.0271 - val_loss: 0.0321
Epoch 87/100
36/36 [==============================] - ETA: 0s - loss: 0.0268
Epoch 87: val_loss did not improve from 0.03123
36/36 [==============================] - 5s 146ms/step - loss: 0.0268 - val_loss: 0.0325
Epoch 88/100
36/36 [==============================] - ETA: 0s - loss: 0.0272
Epoch 88: val_loss did not improve from 0.03123
36/36 [==============================] - 6s 178ms/step - loss: 0.0272 - val_loss: 0.0325
Epoch 89/100
36/36 [==============================] - ETA: 0s - loss: 0.0268
Epoch 89: val_loss did not improve from 0.03123
36/36 [==============================] - 5s 144ms/step - loss: 0.0268 - val_loss: 0.0322
Epoch 90/100
36/36 [==============================] - ETA: 0s - loss: 0.0264
Epoch 90: val_loss did not improve from 0.03123
36/36 [==============================] - 6s 177ms/step - loss: 0.0264 - val_loss: 0.0323
Epoch 91/100
36/36 [==============================] - ETA: 0s - loss: 0.0261
Epoch 91: val_loss did not improve from 0.03123
36/36 [==============================] - 5s 146ms/step - loss: 0.0261 - val_loss: 0.0335
Epoch 92/100
36/36 [==============================] - ETA: 0s - loss: 0.0260
Epoch 92: val_loss did not improve from 0.03123
36/36 [==============================] - 6s 160ms/step - loss: 0.0260 - val_loss: 0.0331
Epoch 93/100
36/36 [==============================] - ETA: 0s - loss: 0.0255
Epoch 93: val_loss did not improve from 0.03123
36/36 [==============================] - 6s 162ms/step - loss: 0.0255 - val_loss: 0.0327
Epoch 94/100
36/36 [==============================] - ETA: 0s - loss: 0.0253
Epoch 94: val_loss did not improve from 0.03123
36/36 [==============================] - 5s 145ms/step - loss: 0.0253 - val_loss: 0.0330
Epoch 95/100
36/36 [==============================] - ETA: 0s - loss: 0.0254
Epoch 95: val_loss did not improve from 0.03123
36/36 [==============================] - 6s 178ms/step - loss: 0.0254 - val_loss: 0.0333
Epoch 96/100
36/36 [==============================] - ETA: 0s - loss: 0.0249
Epoch 96: val_loss did not improve from 0.03123
36/36 [==============================] - 5s 147ms/step - loss: 0.0249 - val_loss: 0.0329
Epoch 97/100
36/36 [==============================] - ETA: 0s - loss: 0.0246
Epoch 97: val_loss did not improve from 0.03123
36/36 [==============================] - 6s 180ms/step - loss: 0.0246 - val_loss: 0.0332
Epoch 98/100
36/36 [==============================] - ETA: 0s - loss: 0.0245
Epoch 98: val_loss did not improve from 0.03123
36/36 [==============================] - 5s 148ms/step - loss: 0.0245 - val_loss: 0.0336
Epoch 99/100
36/36 [==============================] - ETA: 0s - loss: 0.0242
Epoch 99: val_loss did not improve from 0.03123
36/36 [==============================] - 7s 187ms/step - loss: 0.0242 - val_loss: 0.0333
Epoch 100/100
36/36 [==============================] - ETA: 0s - loss: 0.0239
Epoch 100: val_loss did not improve from 0.03123
36/36 [==============================] - 5s 147ms/step - loss: 0.0239 - val_loss: 0.0331

import matplotlib.pyplot as plt  # Package untuk visualisasi data

# Visualisasi model
# Get the training loss from the history
loss = history.history['loss']
# Get the validation loss from the history
val_loss = history.history['val_loss']

# Get the number of epochs
epochs = range(1, len(loss) + 1)

# Plot training loss in green
plt.plot(epochs, loss, 'g', label='Training loss')

# Find the epoch with the best validation loss
best_epoch = val_loss.index(min(val_loss)) + 1

# Plot validation loss in orange
plt.plot(epochs, val_loss, 'orange', label='Validation loss')
# Mark the epoch with the best validation loss
plt.plot(best_epoch, min(val_loss), 'ro', label='Best Validation Loss')

plt.title('LSTM Model 30 Days')
plt.xlabel('Epochs')
plt.ylabel('Loss')
plt.legend()
plt.show()

with open("history.pkl", 'wb') as f:  # Menyimpan history model
    pickle.dump(history, f)

b. 90 Days LSTM

Sama dengan sebelumnya, melainkan kali ini menggunakan 90 hari untuk memprediksi sampah 2024

from keras.callbacks import ModelCheckpoint

model_LSTM_90 = tf.keras.Sequential()
model_LSTM_90.add(tf.keras.layers.LSTM(64, return_sequences=True, input_shape=(90, 3)))
model_LSTM_90.add(tf.keras.layers.LSTM(64, return_sequences=True))
model_LSTM_90.add(tf.keras.layers.LSTM(256, return_sequences=False))
model_LSTM_90.add(tf.keras.layers.Dense(1098, activation='relu'))
model_LSTM_90.add(tf.keras.layers.Reshape((366, 3)))
model_LSTM_90.summary()

Model: "sequential_21"
_________________________________________________________________
 Layer (type)                Output Shape              Param #   
=================================================================
 lstm_43 (LSTM)              (None, 90, 64)            17408     
                                                                 
 lstm_44 (LSTM)              (None, 90, 64)            33024     
                                                                 
 lstm_45 (LSTM)              (None, 256)               328704    
                                                                 
 dense_16 (Dense)            (None, 1098)              282186    
                                                                 
 reshape_13 (Reshape)        (None, 366, 3)            0         
                                                                 
=================================================================
Total params: 661322 (2.52 MB)
Trainable params: 661322 (2.52 MB)
Non-trainable params: 0 (0.00 Byte)
_________________________________________________________________

model_LSTM_90.compile(optimizer='adam', loss='mse')

checkpoint_90 = ModelCheckpoint('best_LSTM_90_model.h5', monitor='val_loss', save_best_only=True, verbose=1)

history_90 = model_LSTM_90.fit(X_90, y_90, epochs=100, validation_split=0.2, callbacks=[checkpoint_90])

Epoch 1/100
35/35 [==============================] - ETA: 0s - loss: 0.1313
Epoch 1: val_loss improved from inf to 0.09376, saving model to best_LSTM_90_model.h5
35/35 [==============================] - 22s 464ms/step - loss: 0.1313 - val_loss: 0.0938
Epoch 2/100

/usr/local/lib/python3.10/dist-packages/keras/src/engine/training.py:3103: UserWarning: You are saving your model as an HDF5 file via `model.save()`. This file format is considered legacy. We recommend using instead the native Keras format, e.g. `model.save('my_model.keras')`.
  saving_api.save_model(

35/35 [==============================] - ETA: 0s - loss: 0.1061
Epoch 2: val_loss improved from 0.09376 to 0.09177, saving model to best_LSTM_90_model.h5
35/35 [==============================] - 16s 457ms/step - loss: 0.1061 - val_loss: 0.0918
Epoch 3/100
35/35 [==============================] - ETA: 0s - loss: 0.1055
Epoch 3: val_loss did not improve from 0.09177
35/35 [==============================] - 16s 454ms/step - loss: 0.1055 - val_loss: 0.0921
Epoch 4/100
35/35 [==============================] - ETA: 0s - loss: 0.1055
Epoch 4: val_loss improved from 0.09177 to 0.09164, saving model to best_LSTM_90_model.h5
35/35 [==============================] - 15s 424ms/step - loss: 0.1055 - val_loss: 0.0916
Epoch 5/100
35/35 [==============================] - ETA: 0s - loss: 0.1053
Epoch 5: val_loss did not improve from 0.09164
35/35 [==============================] - 15s 429ms/step - loss: 0.1053 - val_loss: 0.0918
Epoch 6/100
35/35 [==============================] - ETA: 0s - loss: 0.1053
Epoch 6: val_loss did not improve from 0.09164
35/35 [==============================] - 15s 430ms/step - loss: 0.1053 - val_loss: 0.0919
Epoch 7/100
35/35 [==============================] - ETA: 0s - loss: 0.1050
Epoch 7: val_loss did not improve from 0.09164
35/35 [==============================] - 15s 426ms/step - loss: 0.1050 - val_loss: 0.0917
Epoch 8/100
35/35 [==============================] - ETA: 0s - loss: 0.1047
Epoch 8: val_loss improved from 0.09164 to 0.09135, saving model to best_LSTM_90_model.h5
35/35 [==============================] - 17s 476ms/step - loss: 0.1047 - val_loss: 0.0914
Epoch 9/100
35/35 [==============================] - ETA: 0s - loss: 0.1042
Epoch 9: val_loss improved from 0.09135 to 0.09035, saving model to best_LSTM_90_model.h5
35/35 [==============================] - 15s 424ms/step - loss: 0.1042 - val_loss: 0.0903
Epoch 10/100
35/35 [==============================] - ETA: 0s - loss: 0.1038
Epoch 10: val_loss improved from 0.09035 to 0.08989, saving model to best_LSTM_90_model.h5
35/35 [==============================] - 15s 437ms/step - loss: 0.1038 - val_loss: 0.0899
Epoch 11/100
35/35 [==============================] - ETA: 0s - loss: 0.1030
Epoch 11: val_loss improved from 0.08989 to 0.08916, saving model to best_LSTM_90_model.h5
35/35 [==============================] - 15s 426ms/step - loss: 0.1030 - val_loss: 0.0892
Epoch 12/100
35/35 [==============================] - ETA: 0s - loss: 0.1029
Epoch 12: val_loss did not improve from 0.08916
35/35 [==============================] - 15s 423ms/step - loss: 0.1029 - val_loss: 0.0895
Epoch 13/100
35/35 [==============================] - ETA: 0s - loss: 0.1027
Epoch 13: val_loss did not improve from 0.08916
35/35 [==============================] - 16s 459ms/step - loss: 0.1027 - val_loss: 0.0892
Epoch 14/100
35/35 [==============================] - ETA: 0s - loss: 0.1023
Epoch 14: val_loss improved from 0.08916 to 0.08843, saving model to best_LSTM_90_model.h5
35/35 [==============================] - 16s 456ms/step - loss: 0.1023 - val_loss: 0.0884
Epoch 15/100
35/35 [==============================] - ETA: 0s - loss: 0.1019
Epoch 15: val_loss improved from 0.08843 to 0.08760, saving model to best_LSTM_90_model.h5
35/35 [==============================] - 15s 435ms/step - loss: 0.1019 - val_loss: 0.0876
Epoch 16/100
35/35 [==============================] - ETA: 0s - loss: 0.1009
Epoch 16: val_loss did not improve from 0.08760
35/35 [==============================] - 15s 424ms/step - loss: 0.1009 - val_loss: 0.0880
Epoch 17/100
35/35 [==============================] - ETA: 0s - loss: 0.1006
Epoch 17: val_loss improved from 0.08760 to 0.08698, saving model to best_LSTM_90_model.h5
35/35 [==============================] - 15s 427ms/step - loss: 0.1006 - val_loss: 0.0870
Epoch 18/100
35/35 [==============================] - ETA: 0s - loss: 0.0995
Epoch 18: val_loss did not improve from 0.08698
35/35 [==============================] - 16s 452ms/step - loss: 0.0995 - val_loss: 0.0872
Epoch 19/100
35/35 [==============================] - ETA: 0s - loss: 0.0989
Epoch 19: val_loss improved from 0.08698 to 0.08594, saving model to best_LSTM_90_model.h5
35/35 [==============================] - 15s 430ms/step - loss: 0.0989 - val_loss: 0.0859
Epoch 20/100
35/35 [==============================] - ETA: 0s - loss: 0.0983
Epoch 20: val_loss did not improve from 0.08594
35/35 [==============================] - 15s 425ms/step - loss: 0.0983 - val_loss: 0.0882
Epoch 21/100
35/35 [==============================] - ETA: 0s - loss: 0.0964
Epoch 21: val_loss improved from 0.08594 to 0.08432, saving model to best_LSTM_90_model.h5
35/35 [==============================] - 15s 432ms/step - loss: 0.0964 - val_loss: 0.0843
Epoch 22/100
35/35 [==============================] - ETA: 0s - loss: 0.0944
Epoch 22: val_loss improved from 0.08432 to 0.08331, saving model to best_LSTM_90_model.h5
35/35 [==============================] - 15s 427ms/step - loss: 0.0944 - val_loss: 0.0833
Epoch 23/100
35/35 [==============================] - ETA: 0s - loss: 0.0921
Epoch 23: val_loss improved from 0.08331 to 0.07835, saving model to best_LSTM_90_model.h5
35/35 [==============================] - 16s 457ms/step - loss: 0.0921 - val_loss: 0.0783
Epoch 24/100
35/35 [==============================] - ETA: 0s - loss: 0.0890
Epoch 24: val_loss improved from 0.07835 to 0.07690, saving model to best_LSTM_90_model.h5
35/35 [==============================] - 15s 427ms/step - loss: 0.0890 - val_loss: 0.0769
Epoch 25/100
35/35 [==============================] - ETA: 0s - loss: 0.0858
Epoch 25: val_loss improved from 0.07690 to 0.07363, saving model to best_LSTM_90_model.h5
35/35 [==============================] - 15s 427ms/step - loss: 0.0858 - val_loss: 0.0736
Epoch 26/100
35/35 [==============================] - ETA: 0s - loss: 0.0845
Epoch 26: val_loss improved from 0.07363 to 0.07264, saving model to best_LSTM_90_model.h5
35/35 [==============================] - 15s 425ms/step - loss: 0.0845 - val_loss: 0.0726
Epoch 27/100
35/35 [==============================] - ETA: 0s - loss: 0.0834
Epoch 27: val_loss improved from 0.07264 to 0.07164, saving model to best_LSTM_90_model.h5
35/35 [==============================] - 16s 470ms/step - loss: 0.0834 - val_loss: 0.0716
Epoch 28/100
35/35 [==============================] - ETA: 0s - loss: 0.0826
Epoch 28: val_loss improved from 0.07164 to 0.07027, saving model to best_LSTM_90_model.h5
35/35 [==============================] - 16s 444ms/step - loss: 0.0826 - val_loss: 0.0703
Epoch 29/100
35/35 [==============================] - ETA: 0s - loss: 0.0819
Epoch 29: val_loss did not improve from 0.07027
35/35 [==============================] - 15s 428ms/step - loss: 0.0819 - val_loss: 0.0723
Epoch 30/100
35/35 [==============================] - ETA: 0s - loss: 0.0816
Epoch 30: val_loss improved from 0.07027 to 0.06996, saving model to best_LSTM_90_model.h5
35/35 [==============================] - 15s 428ms/step - loss: 0.0816 - val_loss: 0.0700
Epoch 31/100
35/35 [==============================] - ETA: 0s - loss: 0.0812
Epoch 31: val_loss did not improve from 0.06996
35/35 [==============================] - 15s 426ms/step - loss: 0.0812 - val_loss: 0.0716
Epoch 32/100
35/35 [==============================] - ETA: 0s - loss: 0.0809
Epoch 32: val_loss improved from 0.06996 to 0.06990, saving model to best_LSTM_90_model.h5
35/35 [==============================] - 15s 427ms/step - loss: 0.0809 - val_loss: 0.0699
Epoch 33/100
35/35 [==============================] - ETA: 0s - loss: 0.0807
Epoch 33: val_loss did not improve from 0.06990
35/35 [==============================] - 16s 458ms/step - loss: 0.0807 - val_loss: 0.0708
Epoch 34/100
35/35 [==============================] - ETA: 0s - loss: 0.0803
Epoch 34: val_loss improved from 0.06990 to 0.06892, saving model to best_LSTM_90_model.h5
35/35 [==============================] - 15s 426ms/step - loss: 0.0803 - val_loss: 0.0689
Epoch 35/100
35/35 [==============================] - ETA: 0s - loss: 0.0794
Epoch 35: val_loss did not improve from 0.06892
35/35 [==============================] - 15s 426ms/step - loss: 0.0794 - val_loss: 0.0693
Epoch 36/100
35/35 [==============================] - ETA: 0s - loss: 0.0790
Epoch 36: val_loss improved from 0.06892 to 0.06820, saving model to best_LSTM_90_model.h5
35/35 [==============================] - 15s 424ms/step - loss: 0.0790 - val_loss: 0.0682
Epoch 37/100
35/35 [==============================] - ETA: 0s - loss: 0.0782
Epoch 37: val_loss did not improve from 0.06820
35/35 [==============================] - 15s 424ms/step - loss: 0.0782 - val_loss: 0.0682
Epoch 38/100
35/35 [==============================] - ETA: 0s - loss: 0.0777
Epoch 38: val_loss improved from 0.06820 to 0.06672, saving model to best_LSTM_90_model.h5
35/35 [==============================] - 16s 473ms/step - loss: 0.0777 - val_loss: 0.0667
Epoch 39/100
35/35 [==============================] - ETA: 0s - loss: 0.0764
Epoch 39: val_loss improved from 0.06672 to 0.06540, saving model to best_LSTM_90_model.h5
35/35 [==============================] - 15s 427ms/step - loss: 0.0764 - val_loss: 0.0654
Epoch 40/100
35/35 [==============================] - ETA: 0s - loss: 0.0748
Epoch 40: val_loss improved from 0.06540 to 0.06445, saving model to best_LSTM_90_model.h5
35/35 [==============================] - 16s 458ms/step - loss: 0.0748 - val_loss: 0.0645
Epoch 41/100
35/35 [==============================] - ETA: 0s - loss: 0.0730
Epoch 41: val_loss improved from 0.06445 to 0.06284, saving model to best_LSTM_90_model.h5
35/35 [==============================] - 15s 425ms/step - loss: 0.0730 - val_loss: 0.0628
Epoch 42/100
35/35 [==============================] - ETA: 0s - loss: 0.0720
Epoch 42: val_loss improved from 0.06284 to 0.06253, saving model to best_LSTM_90_model.h5
35/35 [==============================] - 15s 424ms/step - loss: 0.0720 - val_loss: 0.0625
Epoch 43/100
35/35 [==============================] - ETA: 0s - loss: 0.0709
Epoch 43: val_loss improved from 0.06253 to 0.06114, saving model to best_LSTM_90_model.h5
35/35 [==============================] - 15s 434ms/step - loss: 0.0709 - val_loss: 0.0611
Epoch 44/100
35/35 [==============================] - ETA: 0s - loss: 0.0693
Epoch 44: val_loss improved from 0.06114 to 0.05921, saving model to best_LSTM_90_model.h5
35/35 [==============================] - 16s 444ms/step - loss: 0.0693 - val_loss: 0.0592
Epoch 45/100
35/35 [==============================] - ETA: 0s - loss: 0.0675
Epoch 45: val_loss improved from 0.05921 to 0.05770, saving model to best_LSTM_90_model.h5
35/35 [==============================] - 15s 426ms/step - loss: 0.0675 - val_loss: 0.0577
Epoch 46/100
35/35 [==============================] - ETA: 0s - loss: 0.0664
Epoch 46: val_loss improved from 0.05770 to 0.05753, saving model to best_LSTM_90_model.h5
35/35 [==============================] - 15s 427ms/step - loss: 0.0664 - val_loss: 0.0575
Epoch 47/100
35/35 [==============================] - ETA: 0s - loss: 0.0646
Epoch 47: val_loss improved from 0.05753 to 0.05562, saving model to best_LSTM_90_model.h5
35/35 [==============================] - 15s 427ms/step - loss: 0.0646 - val_loss: 0.0556
Epoch 48/100
35/35 [==============================] - ETA: 0s - loss: 0.0628
Epoch 48: val_loss improved from 0.05562 to 0.05508, saving model to best_LSTM_90_model.h5
35/35 [==============================] - 15s 424ms/step - loss: 0.0628 - val_loss: 0.0551
Epoch 49/100
35/35 [==============================] - ETA: 0s - loss: 0.0612
Epoch 49: val_loss improved from 0.05508 to 0.05237, saving model to best_LSTM_90_model.h5
35/35 [==============================] - 16s 459ms/step - loss: 0.0612 - val_loss: 0.0524
Epoch 50/100
35/35 [==============================] - ETA: 0s - loss: 0.0591
Epoch 50: val_loss improved from 0.05237 to 0.05140, saving model to best_LSTM_90_model.h5
35/35 [==============================] - 15s 430ms/step - loss: 0.0591 - val_loss: 0.0514
Epoch 51/100
35/35 [==============================] - ETA: 0s - loss: 0.0579
Epoch 51: val_loss improved from 0.05140 to 0.05084, saving model to best_LSTM_90_model.h5
35/35 [==============================] - 15s 425ms/step - loss: 0.0579 - val_loss: 0.0508
Epoch 52/100
35/35 [==============================] - ETA: 0s - loss: 0.0565
Epoch 52: val_loss improved from 0.05084 to 0.04909, saving model to best_LSTM_90_model.h5
35/35 [==============================] - 15s 428ms/step - loss: 0.0565 - val_loss: 0.0491
Epoch 53/100
35/35 [==============================] - ETA: 0s - loss: 0.0542
Epoch 53: val_loss improved from 0.04909 to 0.04833, saving model to best_LSTM_90_model.h5
35/35 [==============================] - 16s 455ms/step - loss: 0.0542 - val_loss: 0.0483
Epoch 54/100
35/35 [==============================] - ETA: 0s - loss: 0.0521
Epoch 54: val_loss improved from 0.04833 to 0.04566, saving model to best_LSTM_90_model.h5
35/35 [==============================] - 16s 456ms/step - loss: 0.0521 - val_loss: 0.0457
Epoch 55/100
35/35 [==============================] - ETA: 0s - loss: 0.0507
Epoch 55: val_loss improved from 0.04566 to 0.04438, saving model to best_LSTM_90_model.h5
35/35 [==============================] - 15s 427ms/step - loss: 0.0507 - val_loss: 0.0444
Epoch 56/100
35/35 [==============================] - ETA: 0s - loss: 0.0488
Epoch 56: val_loss improved from 0.04438 to 0.04342, saving model to best_LSTM_90_model.h5
35/35 [==============================] - 15s 425ms/step - loss: 0.0488 - val_loss: 0.0434
Epoch 57/100
35/35 [==============================] - ETA: 0s - loss: 0.0470
Epoch 57: val_loss improved from 0.04342 to 0.04226, saving model to best_LSTM_90_model.h5
35/35 [==============================] - 15s 426ms/step - loss: 0.0470 - val_loss: 0.0423
Epoch 58/100
35/35 [==============================] - ETA: 0s - loss: 0.0452
Epoch 58: val_loss improved from 0.04226 to 0.04001, saving model to best_LSTM_90_model.h5
35/35 [==============================] - 15s 429ms/step - loss: 0.0452 - val_loss: 0.0400
Epoch 59/100
35/35 [==============================] - ETA: 0s - loss: 0.0445
Epoch 59: val_loss improved from 0.04001 to 0.03955, saving model to best_LSTM_90_model.h5
35/35 [==============================] - 15s 439ms/step - loss: 0.0445 - val_loss: 0.0396
Epoch 60/100
35/35 [==============================] - ETA: 0s - loss: 0.0434
Epoch 60: val_loss improved from 0.03955 to 0.03912, saving model to best_LSTM_90_model.h5
35/35 [==============================] - 15s 430ms/step - loss: 0.0434 - val_loss: 0.0391
Epoch 61/100
35/35 [==============================] - ETA: 0s - loss: 0.0423
Epoch 61: val_loss improved from 0.03912 to 0.03848, saving model to best_LSTM_90_model.h5
35/35 [==============================] - 15s 425ms/step - loss: 0.0423 - val_loss: 0.0385
Epoch 62/100
35/35 [==============================] - ETA: 0s - loss: 0.0416
Epoch 62: val_loss improved from 0.03848 to 0.03797, saving model to best_LSTM_90_model.h5
35/35 [==============================] - 15s 423ms/step - loss: 0.0416 - val_loss: 0.0380
Epoch 63/100
35/35 [==============================] - ETA: 0s - loss: 0.0403
Epoch 63: val_loss improved from 0.03797 to 0.03718, saving model to best_LSTM_90_model.h5
35/35 [==============================] - 15s 424ms/step - loss: 0.0403 - val_loss: 0.0372
Epoch 64/100
35/35 [==============================] - ETA: 0s - loss: 0.0388
Epoch 64: val_loss improved from 0.03718 to 0.03575, saving model to best_LSTM_90_model.h5
35/35 [==============================] - 15s 424ms/step - loss: 0.0388 - val_loss: 0.0358
Epoch 65/100
35/35 [==============================] - ETA: 0s - loss: 0.0381
Epoch 65: val_loss improved from 0.03575 to 0.03568, saving model to best_LSTM_90_model.h5
35/35 [==============================] - 16s 456ms/step - loss: 0.0381 - val_loss: 0.0357
Epoch 66/100
35/35 [==============================] - ETA: 0s - loss: 0.0377
Epoch 66: val_loss did not improve from 0.03568
35/35 [==============================] - 16s 456ms/step - loss: 0.0377 - val_loss: 0.0359
Epoch 67/100
35/35 [==============================] - ETA: 0s - loss: 0.0380
Epoch 67: val_loss did not improve from 0.03568
35/35 [==============================] - 15s 423ms/step - loss: 0.0380 - val_loss: 0.0363
Epoch 68/100
35/35 [==============================] - ETA: 0s - loss: 0.0380
Epoch 68: val_loss improved from 0.03568 to 0.03468, saving model to best_LSTM_90_model.h5
35/35 [==============================] - 15s 425ms/step - loss: 0.0380 - val_loss: 0.0347
Epoch 69/100
35/35 [==============================] - ETA: 0s - loss: 0.0373
Epoch 69: val_loss improved from 0.03468 to 0.03448, saving model to best_LSTM_90_model.h5
35/35 [==============================] - 15s 430ms/step - loss: 0.0373 - val_loss: 0.0345
Epoch 70/100
35/35 [==============================] - ETA: 0s - loss: 0.0372
Epoch 70: val_loss did not improve from 0.03448
35/35 [==============================] - 16s 454ms/step - loss: 0.0372 - val_loss: 0.0348
Epoch 71/100
35/35 [==============================] - ETA: 0s - loss: 0.0365
Epoch 71: val_loss improved from 0.03448 to 0.03403, saving model to best_LSTM_90_model.h5
35/35 [==============================] - 15s 422ms/step - loss: 0.0365 - val_loss: 0.0340
Epoch 72/100
35/35 [==============================] - ETA: 0s - loss: 0.0357
Epoch 72: val_loss improved from 0.03403 to 0.03378, saving model to best_LSTM_90_model.h5
35/35 [==============================] - 15s 426ms/step - loss: 0.0357 - val_loss: 0.0338
Epoch 73/100
35/35 [==============================] - ETA: 0s - loss: 0.0352
Epoch 73: val_loss improved from 0.03378 to 0.03353, saving model to best_LSTM_90_model.h5
35/35 [==============================] - 15s 432ms/step - loss: 0.0352 - val_loss: 0.0335
Epoch 74/100
35/35 [==============================] - ETA: 0s - loss: 0.0350
Epoch 74: val_loss improved from 0.03353 to 0.03340, saving model to best_LSTM_90_model.h5
35/35 [==============================] - 15s 435ms/step - loss: 0.0350 - val_loss: 0.0334
Epoch 75/100
35/35 [==============================] - ETA: 0s - loss: 0.0346
Epoch 75: val_loss improved from 0.03340 to 0.03320, saving model to best_LSTM_90_model.h5
35/35 [==============================] - 17s 485ms/step - loss: 0.0346 - val_loss: 0.0332
Epoch 76/100
35/35 [==============================] - ETA: 0s - loss: 0.0344
Epoch 76: val_loss improved from 0.03320 to 0.03305, saving model to best_LSTM_90_model.h5
35/35 [==============================] - 15s 428ms/step - loss: 0.0344 - val_loss: 0.0331
Epoch 77/100
35/35 [==============================] - ETA: 0s - loss: 0.0341
Epoch 77: val_loss improved from 0.03305 to 0.03295, saving model to best_LSTM_90_model.h5
35/35 [==============================] - 15s 431ms/step - loss: 0.0341 - val_loss: 0.0330
Epoch 78/100
35/35 [==============================] - ETA: 0s - loss: 0.0337
Epoch 78: val_loss improved from 0.03295 to 0.03271, saving model to best_LSTM_90_model.h5
35/35 [==============================] - 15s 433ms/step - loss: 0.0337 - val_loss: 0.0327
Epoch 79/100
35/35 [==============================] - ETA: 0s - loss: 0.0335
Epoch 79: val_loss improved from 0.03271 to 0.03221, saving model to best_LSTM_90_model.h5
35/35 [==============================] - 18s 510ms/step - loss: 0.0335 - val_loss: 0.0322
Epoch 80/100
35/35 [==============================] - ETA: 0s - loss: 0.0331
Epoch 80: val_loss improved from 0.03221 to 0.03202, saving model to best_LSTM_90_model.h5
35/35 [==============================] - 15s 431ms/step - loss: 0.0331 - val_loss: 0.0320
Epoch 81/100
35/35 [==============================] - ETA: 0s - loss: 0.0329
Epoch 81: val_loss did not improve from 0.03202
35/35 [==============================] - 15s 425ms/step - loss: 0.0329 - val_loss: 0.0321
Epoch 82/100
35/35 [==============================] - ETA: 0s - loss: 0.0327
Epoch 82: val_loss improved from 0.03202 to 0.03197, saving model to best_LSTM_90_model.h5
35/35 [==============================] - 15s 428ms/step - loss: 0.0327 - val_loss: 0.0320
Epoch 83/100
35/35 [==============================] - ETA: 0s - loss: 0.0323
Epoch 83: val_loss improved from 0.03197 to 0.03163, saving model to best_LSTM_90_model.h5
35/35 [==============================] - 15s 428ms/step - loss: 0.0323 - val_loss: 0.0316
Epoch 84/100
35/35 [==============================] - ETA: 0s - loss: 0.0321
Epoch 84: val_loss did not improve from 0.03163
35/35 [==============================] - 17s 485ms/step - loss: 0.0321 - val_loss: 0.0317
Epoch 85/100
35/35 [==============================] - ETA: 0s - loss: 0.0320
Epoch 85: val_loss improved from 0.03163 to 0.03151, saving model to best_LSTM_90_model.h5
35/35 [==============================] - 15s 429ms/step - loss: 0.0320 - val_loss: 0.0315
Epoch 86/100
35/35 [==============================] - ETA: 0s - loss: 0.0320
Epoch 86: val_loss improved from 0.03151 to 0.03144, saving model to best_LSTM_90_model.h5
35/35 [==============================] - 15s 429ms/step - loss: 0.0320 - val_loss: 0.0314
Epoch 87/100
35/35 [==============================] - ETA: 0s - loss: 0.0319
Epoch 87: val_loss improved from 0.03144 to 0.03144, saving model to best_LSTM_90_model.h5
35/35 [==============================] - 15s 427ms/step - loss: 0.0319 - val_loss: 0.0314
Epoch 88/100
35/35 [==============================] - ETA: 0s - loss: 0.0317
Epoch 88: val_loss did not improve from 0.03144
35/35 [==============================] - 15s 428ms/step - loss: 0.0317 - val_loss: 0.0318
Epoch 89/100
35/35 [==============================] - ETA: 0s - loss: 0.0342
Epoch 89: val_loss improved from 0.03144 to 0.03129, saving model to best_LSTM_90_model.h5
35/35 [==============================] - 16s 461ms/step - loss: 0.0342 - val_loss: 0.0313
Epoch 90/100
35/35 [==============================] - ETA: 0s - loss: 0.0314
Epoch 90: val_loss improved from 0.03129 to 0.03073, saving model to best_LSTM_90_model.h5
35/35 [==============================] - 15s 428ms/step - loss: 0.0314 - val_loss: 0.0307
Epoch 91/100
35/35 [==============================] - ETA: 0s - loss: 0.0308
Epoch 91: val_loss did not improve from 0.03073
35/35 [==============================] - 15s 425ms/step - loss: 0.0308 - val_loss: 0.0308
Epoch 92/100
35/35 [==============================] - ETA: 0s - loss: 0.0305
Epoch 92: val_loss improved from 0.03073 to 0.03059, saving model to best_LSTM_90_model.h5
35/35 [==============================] - 16s 469ms/step - loss: 0.0305 - val_loss: 0.0306
Epoch 93/100
35/35 [==============================] - ETA: 0s - loss: 0.0303
Epoch 93: val_loss improved from 0.03059 to 0.03052, saving model to best_LSTM_90_model.h5
35/35 [==============================] - 15s 439ms/step - loss: 0.0303 - val_loss: 0.0305
Epoch 94/100
35/35 [==============================] - ETA: 0s - loss: 0.0301
Epoch 94: val_loss did not improve from 0.03052
35/35 [==============================] - 16s 449ms/step - loss: 0.0301 - val_loss: 0.0306
Epoch 95/100
35/35 [==============================] - ETA: 0s - loss: 0.0300
Epoch 95: val_loss did not improve from 0.03052
35/35 [==============================] - 15s 433ms/step - loss: 0.0300 - val_loss: 0.0306
Epoch 96/100
35/35 [==============================] - ETA: 0s - loss: 0.0300
Epoch 96: val_loss improved from 0.03052 to 0.03034, saving model to best_LSTM_90_model.h5
35/35 [==============================] - 15s 435ms/step - loss: 0.0300 - val_loss: 0.0303
Epoch 97/100
35/35 [==============================] - ETA: 0s - loss: 0.0298
Epoch 97: val_loss did not improve from 0.03034
35/35 [==============================] - 15s 431ms/step - loss: 0.0298 - val_loss: 0.0304
Epoch 98/100
35/35 [==============================] - ETA: 0s - loss: 0.0297
Epoch 98: val_loss improved from 0.03034 to 0.03002, saving model to best_LSTM_90_model.h5
35/35 [==============================] - 15s 444ms/step - loss: 0.0297 - val_loss: 0.0300
Epoch 99/100
35/35 [==============================] - ETA: 0s - loss: 0.0295
Epoch 99: val_loss did not improve from 0.03002
35/35 [==============================] - 15s 439ms/step - loss: 0.0295 - val_loss: 0.0303
Epoch 100/100
35/35 [==============================] - ETA: 0s - loss: 0.0297
Epoch 100: val_loss did not improve from 0.03002
35/35 [==============================] - 17s 476ms/step - loss: 0.0297 - val_loss: 0.0302

# Get the training loss from the history
loss = history_90.history['loss']
# Get the validation loss from the history
val_loss = history_90.history['val_loss']

# Get the number of epochs
epochs = range(1, len(loss) + 1)

# Plot training loss in green
plt.plot(epochs, loss, 'g', label='Training loss')

# Find the epoch with the best validation loss
best_epoch = val_loss.index(min(val_loss)) + 1

# Plot validation loss in orange
plt.plot(epochs, val_loss, 'orange', label='Validation loss')
# Mark the epoch with the best validation loss
plt.plot(best_epoch, min(val_loss), 'ro', label='Best Validation Loss')

plt.title('LSTM Model 90 Days')
plt.xlabel('Epochs')
plt.ylabel('Loss')
plt.legend()
plt.show()

with open("history_90.pkl", 'wb') as f:
    pickle.dump(history_90, f)

c. 366 Days LSTM

Dan percobaan terakhir adalah menggunakan jangka waktu 366 hari untuk memprediksi sampah 2024

model_LSTM_366 = tf.keras.Sequential()
model_LSTM_366.add(tf.keras.layers.LSTM(64, return_sequences=True, input_shape=(366, 3)))
model_LSTM_366.add(tf.keras.layers.LSTM(64, return_sequences=True))
model_LSTM_366.add(tf.keras.layers.LSTM(256, return_sequences=False))
model_LSTM_366.add(tf.keras.layers.Dense(1098, activation='relu'))
model_LSTM_366.add(tf.keras.layers.Reshape((366, 3)))
model_LSTM_366.summary()

Model: "sequential_22"
_________________________________________________________________
 Layer (type)                Output Shape              Param #   
=================================================================
 lstm_46 (LSTM)              (None, 366, 64)           17408     
                                                                 
 lstm_47 (LSTM)              (None, 366, 64)           33024     
                                                                 
 lstm_48 (LSTM)              (None, 256)               328704    
                                                                 
 dense_17 (Dense)            (None, 1098)              282186    
                                                                 
 reshape_14 (Reshape)        (None, 366, 3)            0         
                                                                 
=================================================================
Total params: 661322 (2.52 MB)
Trainable params: 661322 (2.52 MB)
Non-trainable params: 0 (0.00 Byte)
_________________________________________________________________

model_LSTM_366.compile(optimizer='adam', loss='mse')

checkpoint_366 = ModelCheckpoint('best_LSTM_366_model.h5', monitor='val_loss', save_best_only=True, verbose=1)

history_366 = model_LSTM_366.fit(X_366, y_366, epochs=100, validation_split=0.2, callbacks=[checkpoint_366])

Epoch 1/100
28/28 [==============================] - ETA: 0s - loss: 0.1334
Epoch 1: val_loss improved from inf to 0.09401, saving model to best_LSTM_366_model.h5
28/28 [==============================] - 57s 2s/step - loss: 0.1334 - val_loss: 0.0940
Epoch 2/100

/usr/local/lib/python3.10/dist-packages/keras/src/engine/training.py:3103: UserWarning: You are saving your model as an HDF5 file via `model.save()`. This file format is considered legacy. We recommend using instead the native Keras format, e.g. `model.save('my_model.keras')`.
  saving_api.save_model(

28/28 [==============================] - ETA: 0s - loss: 0.1058
Epoch 2: val_loss improved from 0.09401 to 0.09239, saving model to best_LSTM_366_model.h5
28/28 [==============================] - 49s 2s/step - loss: 0.1058 - val_loss: 0.0924
Epoch 3/100
28/28 [==============================] - ETA: 0s - loss: 0.1053
Epoch 3: val_loss improved from 0.09239 to 0.09211, saving model to best_LSTM_366_model.h5
28/28 [==============================] - 49s 2s/step - loss: 0.1053 - val_loss: 0.0921
Epoch 4/100
28/28 [==============================] - ETA: 0s - loss: 0.1052
Epoch 4: val_loss did not improve from 0.09211
28/28 [==============================] - 47s 2s/step - loss: 0.1052 - val_loss: 0.0922
Epoch 5/100
28/28 [==============================] - ETA: 0s - loss: 0.1052
Epoch 5: val_loss improved from 0.09211 to 0.09204, saving model to best_LSTM_366_model.h5
28/28 [==============================] - 50s 2s/step - loss: 0.1052 - val_loss: 0.0920
Epoch 6/100
28/28 [==============================] - ETA: 0s - loss: 0.1052
Epoch 6: val_loss improved from 0.09204 to 0.09194, saving model to best_LSTM_366_model.h5
28/28 [==============================] - 46s 2s/step - loss: 0.1052 - val_loss: 0.0919
Epoch 7/100
28/28 [==============================] - ETA: 0s - loss: 0.1049
Epoch 7: val_loss improved from 0.09194 to 0.09186, saving model to best_LSTM_366_model.h5
28/28 [==============================] - 47s 2s/step - loss: 0.1049 - val_loss: 0.0919
Epoch 8/100
28/28 [==============================] - ETA: 0s - loss: 0.1046
Epoch 8: val_loss improved from 0.09186 to 0.09160, saving model to best_LSTM_366_model.h5
28/28 [==============================] - 48s 2s/step - loss: 0.1046 - val_loss: 0.0916
Epoch 9/100
28/28 [==============================] - ETA: 0s - loss: 0.1040
Epoch 9: val_loss improved from 0.09160 to 0.09071, saving model to best_LSTM_366_model.h5
28/28 [==============================] - 48s 2s/step - loss: 0.1040 - val_loss: 0.0907
Epoch 10/100
28/28 [==============================] - ETA: 0s - loss: 0.1036
Epoch 10: val_loss improved from 0.09071 to 0.09042, saving model to best_LSTM_366_model.h5
28/28 [==============================] - 47s 2s/step - loss: 0.1036 - val_loss: 0.0904
Epoch 11/100
28/28 [==============================] - ETA: 0s - loss: 0.1031
Epoch 11: val_loss improved from 0.09042 to 0.08999, saving model to best_LSTM_366_model.h5
28/28 [==============================] - 48s 2s/step - loss: 0.1031 - val_loss: 0.0900
Epoch 12/100
28/28 [==============================] - ETA: 0s - loss: 0.1022
Epoch 12: val_loss improved from 0.08999 to 0.08920, saving model to best_LSTM_366_model.h5
28/28 [==============================] - 49s 2s/step - loss: 0.1022 - val_loss: 0.0892
Epoch 13/100
28/28 [==============================] - ETA: 0s - loss: 0.1011
Epoch 13: val_loss improved from 0.08920 to 0.08784, saving model to best_LSTM_366_model.h5
28/28 [==============================] - 50s 2s/step - loss: 0.1011 - val_loss: 0.0878
Epoch 14/100
28/28 [==============================] - ETA: 0s - loss: 0.1004
Epoch 14: val_loss improved from 0.08784 to 0.08690, saving model to best_LSTM_366_model.h5
28/28 [==============================] - 46s 2s/step - loss: 0.1004 - val_loss: 0.0869
Epoch 15/100
28/28 [==============================] - ETA: 0s - loss: 0.0990
Epoch 15: val_loss improved from 0.08690 to 0.08628, saving model to best_LSTM_366_model.h5
28/28 [==============================] - 47s 2s/step - loss: 0.0990 - val_loss: 0.0863
Epoch 16/100
28/28 [==============================] - ETA: 0s - loss: 0.0984
Epoch 16: val_loss improved from 0.08628 to 0.08556, saving model to best_LSTM_366_model.h5
28/28 [==============================] - 47s 2s/step - loss: 0.0984 - val_loss: 0.0856
Epoch 17/100
28/28 [==============================] - ETA: 0s - loss: 0.0981
Epoch 17: val_loss did not improve from 0.08556
28/28 [==============================] - 48s 2s/step - loss: 0.0981 - val_loss: 0.0856
Epoch 18/100
28/28 [==============================] - ETA: 0s - loss: 0.0978
Epoch 18: val_loss improved from 0.08556 to 0.08504, saving model to best_LSTM_366_model.h5
28/28 [==============================] - 46s 2s/step - loss: 0.0978 - val_loss: 0.0850
Epoch 19/100
28/28 [==============================] - ETA: 0s - loss: 0.0969
Epoch 19: val_loss improved from 0.08504 to 0.08459, saving model to best_LSTM_366_model.h5
28/28 [==============================] - 47s 2s/step - loss: 0.0969 - val_loss: 0.0846
Epoch 20/100
28/28 [==============================] - ETA: 0s - loss: 0.0967
Epoch 20: val_loss did not improve from 0.08459
28/28 [==============================] - 47s 2s/step - loss: 0.0967 - val_loss: 0.0848
Epoch 21/100
28/28 [==============================] - ETA: 0s - loss: 0.0966
Epoch 21: val_loss improved from 0.08459 to 0.08428, saving model to best_LSTM_366_model.h5
28/28 [==============================] - 46s 2s/step - loss: 0.0966 - val_loss: 0.0843
Epoch 22/100
28/28 [==============================] - ETA: 0s - loss: 0.0967
Epoch 22: val_loss did not improve from 0.08428
28/28 [==============================] - 48s 2s/step - loss: 0.0967 - val_loss: 0.0845
Epoch 23/100
28/28 [==============================] - ETA: 0s - loss: 0.0965
Epoch 23: val_loss improved from 0.08428 to 0.08408, saving model to best_LSTM_366_model.h5
28/28 [==============================] - 47s 2s/step - loss: 0.0965 - val_loss: 0.0841
Epoch 24/100
28/28 [==============================] - ETA: 0s - loss: 0.0963
Epoch 24: val_loss improved from 0.08408 to 0.08390, saving model to best_LSTM_366_model.h5
28/28 [==============================] - 46s 2s/step - loss: 0.0963 - val_loss: 0.0839
Epoch 25/100
28/28 [==============================] - ETA: 0s - loss: 0.0959
Epoch 25: val_loss improved from 0.08390 to 0.08358, saving model to best_LSTM_366_model.h5
28/28 [==============================] - 48s 2s/step - loss: 0.0959 - val_loss: 0.0836
Epoch 26/100
28/28 [==============================] - ETA: 0s - loss: 0.0956
Epoch 26: val_loss did not improve from 0.08358
28/28 [==============================] - 48s 2s/step - loss: 0.0956 - val_loss: 0.0837
Epoch 27/100
28/28 [==============================] - ETA: 0s - loss: 0.0954
Epoch 27: val_loss improved from 0.08358 to 0.08263, saving model to best_LSTM_366_model.h5
28/28 [==============================] - 47s 2s/step - loss: 0.0954 - val_loss: 0.0826
Epoch 28/100
28/28 [==============================] - ETA: 0s - loss: 0.0951
Epoch 28: val_loss improved from 0.08263 to 0.08257, saving model to best_LSTM_366_model.h5
28/28 [==============================] - 46s 2s/step - loss: 0.0951 - val_loss: 0.0826
Epoch 29/100
28/28 [==============================] - ETA: 0s - loss: 0.0944
Epoch 29: val_loss improved from 0.08257 to 0.08205, saving model to best_LSTM_366_model.h5
28/28 [==============================] - 47s 2s/step - loss: 0.0944 - val_loss: 0.0820
Epoch 30/100
28/28 [==============================] - ETA: 0s - loss: 0.0944
Epoch 30: val_loss did not improve from 0.08205
28/28 [==============================] - 48s 2s/step - loss: 0.0944 - val_loss: 0.0821
Epoch 31/100
28/28 [==============================] - ETA: 0s - loss: 0.0939
Epoch 31: val_loss improved from 0.08205 to 0.08178, saving model to best_LSTM_366_model.h5
28/28 [==============================] - 47s 2s/step - loss: 0.0939 - val_loss: 0.0818
Epoch 32/100
28/28 [==============================] - ETA: 0s - loss: 0.0938
Epoch 32: val_loss did not improve from 0.08178
28/28 [==============================] - 47s 2s/step - loss: 0.0938 - val_loss: 0.0819
Epoch 33/100
28/28 [==============================] - ETA: 0s - loss: 0.0937
Epoch 33: val_loss did not improve from 0.08178
28/28 [==============================] - 47s 2s/step - loss: 0.0937 - val_loss: 0.0818
Epoch 34/100
28/28 [==============================] - ETA: 0s - loss: 0.0937
Epoch 34: val_loss did not improve from 0.08178
28/28 [==============================] - 48s 2s/step - loss: 0.0937 - val_loss: 0.0821
Epoch 35/100
28/28 [==============================] - ETA: 0s - loss: 0.0938
Epoch 35: val_loss improved from 0.08178 to 0.08143, saving model to best_LSTM_366_model.h5
28/28 [==============================] - 47s 2s/step - loss: 0.0938 - val_loss: 0.0814
Epoch 36/100
28/28 [==============================] - ETA: 0s - loss: 0.0934
Epoch 36: val_loss improved from 0.08143 to 0.08120, saving model to best_LSTM_366_model.h5
28/28 [==============================] - 48s 2s/step - loss: 0.0934 - val_loss: 0.0812
Epoch 37/100
28/28 [==============================] - ETA: 0s - loss: 0.0932
Epoch 37: val_loss did not improve from 0.08120
28/28 [==============================] - 51s 2s/step - loss: 0.0932 - val_loss: 0.0814
Epoch 38/100
28/28 [==============================] - ETA: 0s - loss: 0.0931
Epoch 38: val_loss improved from 0.08120 to 0.08118, saving model to best_LSTM_366_model.h5
28/28 [==============================] - 52s 2s/step - loss: 0.0931 - val_loss: 0.0812
Epoch 39/100
28/28 [==============================] - ETA: 0s - loss: 0.0930
Epoch 39: val_loss improved from 0.08118 to 0.08093, saving model to best_LSTM_366_model.h5
28/28 [==============================] - 49s 2s/step - loss: 0.0930 - val_loss: 0.0809
Epoch 40/100
28/28 [==============================] - ETA: 0s - loss: 0.0926
Epoch 40: val_loss improved from 0.08093 to 0.08046, saving model to best_LSTM_366_model.h5
28/28 [==============================] - 49s 2s/step - loss: 0.0926 - val_loss: 0.0805
Epoch 41/100
28/28 [==============================] - ETA: 0s - loss: 0.0918
Epoch 41: val_loss improved from 0.08046 to 0.07967, saving model to best_LSTM_366_model.h5
28/28 [==============================] - 49s 2s/step - loss: 0.0918 - val_loss: 0.0797
Epoch 42/100
28/28 [==============================] - ETA: 0s - loss: 0.0913
Epoch 42: val_loss did not improve from 0.07967
28/28 [==============================] - 50s 2s/step - loss: 0.0913 - val_loss: 0.0797
Epoch 43/100
28/28 [==============================] - ETA: 0s - loss: 0.0906
Epoch 43: val_loss improved from 0.07967 to 0.07913, saving model to best_LSTM_366_model.h5
28/28 [==============================] - 50s 2s/step - loss: 0.0906 - val_loss: 0.0791
Epoch 44/100
28/28 [==============================] - ETA: 0s - loss: 0.0907
Epoch 44: val_loss improved from 0.07913 to 0.07891, saving model to best_LSTM_366_model.h5
28/28 [==============================] - 47s 2s/step - loss: 0.0907 - val_loss: 0.0789
Epoch 45/100
28/28 [==============================] - ETA: 0s - loss: 0.0902
Epoch 45: val_loss improved from 0.07891 to 0.07869, saving model to best_LSTM_366_model.h5
28/28 [==============================] - 48s 2s/step - loss: 0.0902 - val_loss: 0.0787
Epoch 46/100
28/28 [==============================] - ETA: 0s - loss: 0.0900
Epoch 46: val_loss improved from 0.07869 to 0.07856, saving model to best_LSTM_366_model.h5
28/28 [==============================] - 50s 2s/step - loss: 0.0900 - val_loss: 0.0786
Epoch 47/100
28/28 [==============================] - ETA: 0s - loss: 0.0892
Epoch 47: val_loss improved from 0.07856 to 0.07807, saving model to best_LSTM_366_model.h5
28/28 [==============================] - 49s 2s/step - loss: 0.0892 - val_loss: 0.0781
Epoch 48/100
28/28 [==============================] - ETA: 0s - loss: 0.0888
Epoch 48: val_loss improved from 0.07807 to 0.07735, saving model to best_LSTM_366_model.h5
28/28 [==============================] - 48s 2s/step - loss: 0.0888 - val_loss: 0.0773
Epoch 49/100
28/28 [==============================] - ETA: 0s - loss: 0.0880
Epoch 49: val_loss improved from 0.07735 to 0.07721, saving model to best_LSTM_366_model.h5
28/28 [==============================] - 47s 2s/step - loss: 0.0880 - val_loss: 0.0772
Epoch 50/100
28/28 [==============================] - ETA: 0s - loss: 0.0879
Epoch 50: val_loss did not improve from 0.07721
28/28 [==============================] - 49s 2s/step - loss: 0.0879 - val_loss: 0.0780
Epoch 51/100
28/28 [==============================] - ETA: 0s - loss: 0.0875
Epoch 51: val_loss improved from 0.07721 to 0.07489, saving model to best_LSTM_366_model.h5
28/28 [==============================] - 48s 2s/step - loss: 0.0875 - val_loss: 0.0749
Epoch 52/100
28/28 [==============================] - ETA: 0s - loss: 0.0853
Epoch 52: val_loss improved from 0.07489 to 0.07343, saving model to best_LSTM_366_model.h5
28/28 [==============================] - 48s 2s/step - loss: 0.0853 - val_loss: 0.0734
Epoch 53/100
28/28 [==============================] - ETA: 0s - loss: 0.0847
Epoch 53: val_loss improved from 0.07343 to 0.07311, saving model to best_LSTM_366_model.h5
28/28 [==============================] - 47s 2s/step - loss: 0.0847 - val_loss: 0.0731
Epoch 54/100
28/28 [==============================] - ETA: 0s - loss: 0.0845
Epoch 54: val_loss improved from 0.07311 to 0.07271, saving model to best_LSTM_366_model.h5
28/28 [==============================] - 46s 2s/step - loss: 0.0845 - val_loss: 0.0727
Epoch 55/100
28/28 [==============================] - ETA: 0s - loss: 0.0843
Epoch 55: val_loss improved from 0.07271 to 0.07270, saving model to best_LSTM_366_model.h5
28/28 [==============================] - 48s 2s/step - loss: 0.0843 - val_loss: 0.0727
Epoch 56/100
28/28 [==============================] - ETA: 0s - loss: 0.0842
Epoch 56: val_loss improved from 0.07270 to 0.07257, saving model to best_LSTM_366_model.h5
28/28 [==============================] - 48s 2s/step - loss: 0.0842 - val_loss: 0.0726
Epoch 57/100
28/28 [==============================] - ETA: 0s - loss: 0.0840
Epoch 57: val_loss did not improve from 0.07257
28/28 [==============================] - 47s 2s/step - loss: 0.0840 - val_loss: 0.0726
Epoch 58/100
28/28 [==============================] - ETA: 0s - loss: 0.0837
Epoch 58: val_loss did not improve from 0.07257
28/28 [==============================] - 48s 2s/step - loss: 0.0837 - val_loss: 0.0727
Epoch 59/100
28/28 [==============================] - ETA: 0s - loss: 0.0834
Epoch 59: val_loss improved from 0.07257 to 0.07246, saving model to best_LSTM_366_model.h5
28/28 [==============================] - 50s 2s/step - loss: 0.0834 - val_loss: 0.0725
Epoch 60/100
28/28 [==============================] - ETA: 0s - loss: 0.0839
Epoch 60: val_loss did not improve from 0.07246
28/28 [==============================] - 47s 2s/step - loss: 0.0839 - val_loss: 0.0732
Epoch 61/100
28/28 [==============================] - ETA: 0s - loss: 0.0829
Epoch 61: val_loss did not improve from 0.07246
28/28 [==============================] - 47s 2s/step - loss: 0.0829 - val_loss: 0.0734
Epoch 62/100
28/28 [==============================] - ETA: 0s - loss: 0.0823
Epoch 62: val_loss improved from 0.07246 to 0.07175, saving model to best_LSTM_366_model.h5
28/28 [==============================] - 47s 2s/step - loss: 0.0823 - val_loss: 0.0717
Epoch 63/100
28/28 [==============================] - ETA: 0s - loss: 0.0816
Epoch 63: val_loss did not improve from 0.07175
28/28 [==============================] - 48s 2s/step - loss: 0.0816 - val_loss: 0.0725
Epoch 64/100
28/28 [==============================] - ETA: 0s - loss: 0.0810
Epoch 64: val_loss improved from 0.07175 to 0.07145, saving model to best_LSTM_366_model.h5
28/28 [==============================] - 47s 2s/step - loss: 0.0810 - val_loss: 0.0715
Epoch 65/100
28/28 [==============================] - ETA: 0s - loss: 0.0805
Epoch 65: val_loss did not improve from 0.07145
28/28 [==============================] - 47s 2s/step - loss: 0.0805 - val_loss: 0.0727
Epoch 66/100
28/28 [==============================] - ETA: 0s - loss: 0.0800
Epoch 66: val_loss did not improve from 0.07145
28/28 [==============================] - 47s 2s/step - loss: 0.0800 - val_loss: 0.0715
Epoch 67/100
28/28 [==============================] - ETA: 0s - loss: 0.0783
Epoch 67: val_loss improved from 0.07145 to 0.07033, saving model to best_LSTM_366_model.h5
28/28 [==============================] - 48s 2s/step - loss: 0.0783 - val_loss: 0.0703
Epoch 68/100
28/28 [==============================] - ETA: 0s - loss: 0.0762
Epoch 68: val_loss improved from 0.07033 to 0.06623, saving model to best_LSTM_366_model.h5
28/28 [==============================] - 47s 2s/step - loss: 0.0762 - val_loss: 0.0662
Epoch 69/100
28/28 [==============================] - ETA: 0s - loss: 0.0735
Epoch 69: val_loss improved from 0.06623 to 0.06404, saving model to best_LSTM_366_model.h5
28/28 [==============================] - 48s 2s/step - loss: 0.0735 - val_loss: 0.0640
Epoch 70/100
28/28 [==============================] - ETA: 0s - loss: 0.0707
Epoch 70: val_loss improved from 0.06404 to 0.06235, saving model to best_LSTM_366_model.h5
28/28 [==============================] - 49s 2s/step - loss: 0.0707 - val_loss: 0.0624
Epoch 71/100
28/28 [==============================] - ETA: 0s - loss: 0.0670
Epoch 71: val_loss improved from 0.06235 to 0.05952, saving model to best_LSTM_366_model.h5
28/28 [==============================] - 51s 2s/step - loss: 0.0670 - val_loss: 0.0595
Epoch 72/100
28/28 [==============================] - ETA: 0s - loss: 0.0656
Epoch 72: val_loss improved from 0.05952 to 0.05916, saving model to best_LSTM_366_model.h5
28/28 [==============================] - 51s 2s/step - loss: 0.0656 - val_loss: 0.0592
Epoch 73/100
28/28 [==============================] - ETA: 0s - loss: 0.0628
Epoch 73: val_loss improved from 0.05916 to 0.05530, saving model to best_LSTM_366_model.h5
28/28 [==============================] - 49s 2s/step - loss: 0.0628 - val_loss: 0.0553
Epoch 74/100
28/28 [==============================] - ETA: 0s - loss: 0.0596
Epoch 74: val_loss improved from 0.05530 to 0.05269, saving model to best_LSTM_366_model.h5
28/28 [==============================] - 50s 2s/step - loss: 0.0596 - val_loss: 0.0527
Epoch 75/100
28/28 [==============================] - ETA: 0s - loss: 0.0567
Epoch 75: val_loss improved from 0.05269 to 0.05172, saving model to best_LSTM_366_model.h5
28/28 [==============================] - 51s 2s/step - loss: 0.0567 - val_loss: 0.0517
Epoch 76/100
28/28 [==============================] - ETA: 0s - loss: 0.0551
Epoch 76: val_loss improved from 0.05172 to 0.05007, saving model to best_LSTM_366_model.h5
28/28 [==============================] - 49s 2s/step - loss: 0.0551 - val_loss: 0.0501
Epoch 77/100
28/28 [==============================] - ETA: 0s - loss: 0.0537
Epoch 77: val_loss improved from 0.05007 to 0.04817, saving model to best_LSTM_366_model.h5
28/28 [==============================] - 49s 2s/step - loss: 0.0537 - val_loss: 0.0482
Epoch 78/100
28/28 [==============================] - ETA: 0s - loss: 0.0517
Epoch 78: val_loss improved from 0.04817 to 0.04622, saving model to best_LSTM_366_model.h5
28/28 [==============================] - 49s 2s/step - loss: 0.0517 - val_loss: 0.0462
Epoch 79/100
28/28 [==============================] - ETA: 0s - loss: 0.0483
Epoch 79: val_loss improved from 0.04622 to 0.04296, saving model to best_LSTM_366_model.h5
28/28 [==============================] - 49s 2s/step - loss: 0.0483 - val_loss: 0.0430
Epoch 80/100
28/28 [==============================] - ETA: 0s - loss: 0.0451
Epoch 80: val_loss improved from 0.04296 to 0.04102, saving model to best_LSTM_366_model.h5
28/28 [==============================] - 46s 2s/step - loss: 0.0451 - val_loss: 0.0410
Epoch 81/100
28/28 [==============================] - ETA: 0s - loss: 0.0438
Epoch 81: val_loss improved from 0.04102 to 0.03986, saving model to best_LSTM_366_model.h5
28/28 [==============================] - 47s 2s/step - loss: 0.0438 - val_loss: 0.0399
Epoch 82/100
28/28 [==============================] - ETA: 0s - loss: 0.0421
Epoch 82: val_loss improved from 0.03986 to 0.03965, saving model to best_LSTM_366_model.h5
28/28 [==============================] - 47s 2s/step - loss: 0.0421 - val_loss: 0.0396
Epoch 83/100
28/28 [==============================] - ETA: 0s - loss: 0.0407
Epoch 83: val_loss improved from 0.03965 to 0.03798, saving model to best_LSTM_366_model.h5
28/28 [==============================] - 48s 2s/step - loss: 0.0407 - val_loss: 0.0380
Epoch 84/100
28/28 [==============================] - ETA: 0s - loss: 0.0395
Epoch 84: val_loss improved from 0.03798 to 0.03726, saving model to best_LSTM_366_model.h5
28/28 [==============================] - 47s 2s/step - loss: 0.0395 - val_loss: 0.0373
Epoch 85/100
28/28 [==============================] - ETA: 0s - loss: 0.0387
Epoch 85: val_loss improved from 0.03726 to 0.03608, saving model to best_LSTM_366_model.h5
28/28 [==============================] - 46s 2s/step - loss: 0.0387 - val_loss: 0.0361
Epoch 86/100
28/28 [==============================] - ETA: 0s - loss: 0.0377
Epoch 86: val_loss improved from 0.03608 to 0.03587, saving model to best_LSTM_366_model.h5
28/28 [==============================] - 47s 2s/step - loss: 0.0377 - val_loss: 0.0359
Epoch 87/100
28/28 [==============================] - ETA: 0s - loss: 0.0370
Epoch 87: val_loss improved from 0.03587 to 0.03539, saving model to best_LSTM_366_model.h5
28/28 [==============================] - 48s 2s/step - loss: 0.0370 - val_loss: 0.0354
Epoch 88/100
28/28 [==============================] - ETA: 0s - loss: 0.0368
Epoch 88: val_loss did not improve from 0.03539
28/28 [==============================] - 47s 2s/step - loss: 0.0368 - val_loss: 0.0354
Epoch 89/100
28/28 [==============================] - ETA: 0s - loss: 0.0365
Epoch 89: val_loss improved from 0.03539 to 0.03494, saving model to best_LSTM_366_model.h5
28/28 [==============================] - 46s 2s/step - loss: 0.0365 - val_loss: 0.0349
Epoch 90/100
28/28 [==============================] - ETA: 0s - loss: 0.0359
Epoch 90: val_loss improved from 0.03494 to 0.03464, saving model to best_LSTM_366_model.h5
28/28 [==============================] - 47s 2s/step - loss: 0.0359 - val_loss: 0.0346
Epoch 91/100
28/28 [==============================] - ETA: 0s - loss: 0.0355
Epoch 91: val_loss improved from 0.03464 to 0.03402, saving model to best_LSTM_366_model.h5
28/28 [==============================] - 48s 2s/step - loss: 0.0355 - val_loss: 0.0340
Epoch 92/100
28/28 [==============================] - ETA: 0s - loss: 0.0351
Epoch 92: val_loss improved from 0.03402 to 0.03351, saving model to best_LSTM_366_model.h5
28/28 [==============================] - 47s 2s/step - loss: 0.0351 - val_loss: 0.0335
Epoch 93/100
28/28 [==============================] - ETA: 0s - loss: 0.0342
Epoch 93: val_loss improved from 0.03351 to 0.03337, saving model to best_LSTM_366_model.h5
28/28 [==============================] - 48s 2s/step - loss: 0.0342 - val_loss: 0.0334
Epoch 94/100
28/28 [==============================] - ETA: 0s - loss: 0.0340
Epoch 94: val_loss improved from 0.03337 to 0.03317, saving model to best_LSTM_366_model.h5
28/28 [==============================] - 48s 2s/step - loss: 0.0340 - val_loss: 0.0332
Epoch 95/100
28/28 [==============================] - ETA: 0s - loss: 0.0337
Epoch 95: val_loss did not improve from 0.03317
28/28 [==============================] - 46s 2s/step - loss: 0.0337 - val_loss: 0.0333
Epoch 96/100
28/28 [==============================] - ETA: 0s - loss: 0.0337
Epoch 96: val_loss improved from 0.03317 to 0.03310, saving model to best_LSTM_366_model.h5
28/28 [==============================] - 48s 2s/step - loss: 0.0337 - val_loss: 0.0331
Epoch 97/100
28/28 [==============================] - ETA: 0s - loss: 0.0336
Epoch 97: val_loss did not improve from 0.03310
28/28 [==============================] - 47s 2s/step - loss: 0.0336 - val_loss: 0.0334
Epoch 98/100
28/28 [==============================] - ETA: 0s - loss: 0.0331
Epoch 98: val_loss improved from 0.03310 to 0.03265, saving model to best_LSTM_366_model.h5
28/28 [==============================] - 47s 2s/step - loss: 0.0331 - val_loss: 0.0327
Epoch 99/100
28/28 [==============================] - ETA: 0s - loss: 0.0329
Epoch 99: val_loss did not improve from 0.03265
28/28 [==============================] - 46s 2s/step - loss: 0.0329 - val_loss: 0.0329
Epoch 100/100
28/28 [==============================] - ETA: 0s - loss: 0.0324
Epoch 100: val_loss improved from 0.03265 to 0.03227, saving model to best_LSTM_366_model.h5
28/28 [==============================] - 48s 2s/step - loss: 0.0324 - val_loss: 0.0323

# Get the training loss from the history
loss = history_366.history['loss']
# Get the validation loss from the history
val_loss = history_366.history['val_loss']

# Get the number of epochs
epochs = range(1, len(loss) + 1)

# Plot training loss in green
plt.plot(epochs, loss, 'g', label='Training loss')

# Find the epoch with the best validation loss
best_epoch = val_loss.index(min(val_loss)) + 1

# Plot validation loss in orange
plt.plot(epochs, val_loss, 'orange', label='Validation loss')
# Mark the epoch with the best validation loss
plt.plot(best_epoch, min(val_loss), 'ro', label='Best Validation Loss')

plt.title('LSTM Model 366 Days')
plt.xlabel('Epochs')
plt.ylabel('Loss')
plt.legend()
plt.show()

with open("history_366.pkl", 'wb') as f:
    pickle.dump(history_366, f)

3. Prediction (Diganti nama modelnya)

Setelah dilakukan tiga kali percobaan, ditemukan percobaan terbaik adalah percobaan dengan 30 hari (percobaan pertama) ditandai dengan loss validation terkecil yaitu 0.03123

Bagian terakhir adalah memprediksi sampah 2024 dengan model yang telah didapat

X_pred = np.array([df_timeseries[['magelang_utara', 'magelang_tengah', 'magelang_selatan']].iloc[-30:].values])  # Membuat data yaitu 30 hari terakhir untuk diprediksi
y_pred = model_LSTM.predict(X_pred)  # Prediksi menggunakan model terbaik yaitu model 30 Hari
print(y_pred.shape)
print(y_pred[:5])

1/1 [==============================] - 4s 4s/step
(1, 366, 3)
[[[0.5976804  0.4909475  0.7239353 ]
  [0.55852544 0.43864074 0.693532  ]
  [0.5564108  0.45952934 0.65807575]
  ...
  [0.52366143 0.40401322 0.51513886]
  [0.55716836 0.43663803 0.5384773 ]
  [0.57083505 0.3644834  0.47720507]]]

y_pred_denorm = scaler.inverse_transform(y_pred[0])  # Mengembalikan data prediksi yang dinormalisasi menjadi bentuk semula

df_pred = pd.DataFrame(y_pred_denorm, columns=['magelang_utara', 'magelang_tengah', 'magelang_selatan'])  # Membentuk data prediksi menjadi dataframe
df_pred.head()

	magelang_utara	magelang_tengah	magelang_selatan
0	24533.755859	11401.963867	22240.269531
1	22942.888672	10214.337891	21323.457031
2	22856.970703	10688.613281	20254.273438
3	20227.412109	10738.855469	18646.232422
4	18266.132812	9440.942383	17265.660156

# Mengembalikan nilai ke bentuk semula sebelum normalisasi
df_timeseries_denorm = denormalize_dataframe(df_timeseries, ['magelang_utara', 'magelang_tengah', 'magelang_selatan'], scaler)  # Mengembalikan dataframe (bukan data prediksi) yang dinormalisasi menjadi bentuk semula
df_timeseries_denorm.head()

	tanggal	magelang_utara	magelang_tengah	magelang_selatan
0	2019-01-03	7320.0	280.0	3920.0
1	2019-01-04	26154.0	5510.0	4130.0
2	2019-01-05	22700.0	4680.0	5640.0
3	2019-01-06	24320.0	4370.0	8950.0
4	2019-01-07	34742.0	15350.0	16360.0

# Concatenate original_df with pred_df
extended_df = pd.concat([df_timeseries_denorm, df_pred], axis=0)  # Menggabungkan data asli dengan data prediksi

# Extend the date by 366 days
last_date = df_timeseries_denorm['tanggal'].iloc[-1]  # Memperpanjang tanggal satu tahun lebih (2024)
extended_dates = pd.date_range(start=last_date, periods=366)

# Add the extended dates to the DataFrame
extended_df['tanggal'] = pd.concat([df_timeseries_denorm['tanggal'], pd.Series(extended_dates)])  # Menggabungkan tanggal ke dalam dataframe

# Reset index
extended_df.reset_index(drop=True, inplace=True)  # Reset index untuk dataframe

temp_type = np.array([0]*len(extended_df))  # Menambahkan kolom 'type', dimana 0 untuk data asli, 1 untuk data prediksi
temp_type[-366:] = 1

extended_df['type'] = temp_type

extended_df.info()

<class 'pandas.core.frame.DataFrame'>
RangeIndex: 2190 entries, 0 to 2189
Data columns (total 5 columns):
 #   Column            Non-Null Count  Dtype         
---  ------            --------------  -----         
 0   tanggal           2190 non-null   datetime64[ns]
 1   magelang_utara    2190 non-null   float64       
 2   magelang_tengah   2190 non-null   float64       
 3   magelang_selatan  2190 non-null   float64       
 4   type              2190 non-null   int64         
dtypes: datetime64[ns](1), float64(3), int64(1)
memory usage: 85.7 KB

extended_df.to_csv('extended_df.csv', index=False)  # Menyimpan data ke dalam bentuk .csv

import matplotlib.pyplot as plt  # Visualisasi data gabungan

# Assuming date_index_df contains the extended DataFrame with columns 'date', 'value1', 'value2', 'value3', 'pred_value1', 'pred_value2', 'pred_value3'

# Set the 'date' column as the index
date_index_df = extended_df.set_index('tanggal')

# Plot the values
plt.figure(figsize=(10, 6))

plt.plot(date_index_df.index, date_index_df['magelang_utara'], label='Original Value 1')
plt.plot(date_index_df.index, date_index_df['magelang_tengah'], label='Original Value 2')
plt.plot(date_index_df.index, date_index_df['magelang_selatan'], label='Original Value 3')

plt.xlabel('Date')
plt.ylabel('Value')
plt.title('Original and Predicted Values Over Time')
plt.legend()
plt.grid(True)
plt.show()

Powered by Pristi Sukmasetya 2024