初心者のRNN(LSTM) | Kerasで試してみる
上記記事ののコードを実行してみる

from tensorflow.keras.models import Sequential
from tensorflow.keras.layers import Dense, Activation
from tensorflow.keras.layers import Bidirectional
from tensorflow.keras.optimizers import Adam
from tensorflow.keras.callbacks import EarlyStopping
import numpy as np
import matplotlib.pyplot as plt

from tensorflow.keras.layers import LSTM

def sin(x, T=100):
    return np.sin(2.0 * np.pi * x / T)

# sin波にノイズを付与する
def toy_problem(T=100, ampl=0.05):
    x = np.arange(0, 2 * T + 1)
    noise = ampl * np.random.uniform(low=-1.0, high=1.0, size=len(x))
    return sin(x) + noise

f = toy_problem()

def make_dataset(low_data, n_prev=100):

    data, target = [], []
    maxlen = 25

    for i in range(len(low_data)-maxlen):
        data.append(low_data[i:i + maxlen])
        target.append(low_data[i + maxlen])

    re_data = np.array(data).reshape(len(data), maxlen, 1)
    re_target = np.array(target).reshape(len(data), 1)

    return re_data, re_target


#g -> 学習データ，h -> 学習ラベル
g, h = make_dataset(f)

# モデル構築

# 1つの学習データのStep数(今回は25)
length_of_sequence = g.shape[1] 
in_out_neurons = 1
n_hidden = 300

model = Sequential()
model.add(Bidirectional(LSTM(n_hidden, batch_input_shape=(None, length_of_sequence, in_out_neurons), return_sequences=False)))#【コメント】BidirectionalつけるだけでOK
model.add(Dense(in_out_neurons))
model.add(Activation("linear"))
optimizer = Adam(learning_rate=0.001)
model.compile(loss="mean_squared_error", optimizer=optimizer)

early_stopping = EarlyStopping(monitor='val_loss', mode='auto', patience=20)
model.fit(g, h,
          batch_size=300,
          epochs=100,
          validation_split=0.1,
          callbacks=[early_stopping]
          )
#【コメント】early_stoppingにより38epochで停止

Epoch 1/100
1/1 [==============================] - 6s 6s/step - loss: 0.5961 - val_loss: 0.1524
Epoch 2/100
1/1 [==============================] - 1s 520ms/step - loss: 0.3491 - val_loss: 0.0599
Epoch 3/100
1/1 [==============================] - 1s 519ms/step - loss: 0.2150 - val_loss: 0.2243
Epoch 4/100
1/1 [==============================] - 1s 515ms/step - loss: 0.1984 - val_loss: 0.3404
Epoch 5/100
1/1 [==============================] - 1s 529ms/step - loss: 0.1875 - val_loss: 0.2224
Epoch 6/100
1/1 [==============================] - 1s 537ms/step - loss: 0.1272 - val_loss: 0.0953
Epoch 7/100
1/1 [==============================] - 1s 538ms/step - loss: 0.0886 - val_loss: 0.0361
Epoch 8/100
1/1 [==============================] - 1s 532ms/step - loss: 0.0791 - val_loss: 0.0220
Epoch 9/100
1/1 [==============================] - 1s 532ms/step - loss: 0.0765 - val_loss: 0.0217
Epoch 10/100
1/1 [==============================] - 1s 543ms/step - loss: 0.0648 - val_loss: 0.0190
Epoch 11/100
1/1 [==============================] - 1s 548ms/step - loss: 0.0416 - val_loss: 0.0122
Epoch 12/100
1/1 [==============================] - 1s 538ms/step - loss: 0.0178 - val_loss: 0.0081
Epoch 13/100
1/1 [==============================] - 1s 519ms/step - loss: 0.0152 - val_loss: 0.0092
Epoch 14/100
1/1 [==============================] - 1s 551ms/step - loss: 0.0261 - val_loss: 0.0048
Epoch 15/100
1/1 [==============================] - 1s 535ms/step - loss: 0.0114 - val_loss: 0.0030
Epoch 16/100
1/1 [==============================] - 1s 535ms/step - loss: 0.0038 - val_loss: 0.0086
Epoch 17/100
1/1 [==============================] - 1s 523ms/step - loss: 0.0122 - val_loss: 0.0111
Epoch 18/100
1/1 [==============================] - 1s 529ms/step - loss: 0.0165 - val_loss: 0.0062
Epoch 19/100
1/1 [==============================] - 1s 528ms/step - loss: 0.0115 - val_loss: 0.0015
Epoch 20/100
1/1 [==============================] - 1s 544ms/step - loss: 0.0050 - val_loss: 0.0029
Epoch 21/100
1/1 [==============================] - 1s 517ms/step - loss: 0.0032 - val_loss: 0.0065
Epoch 22/100
1/1 [==============================] - 1s 539ms/step - loss: 0.0046 - val_loss: 0.0055
Epoch 23/100
1/1 [==============================] - 1s 528ms/step - loss: 0.0050 - val_loss: 0.0022
Epoch 24/100
1/1 [==============================] - 1s 545ms/step - loss: 0.0040 - val_loss: 0.0015
Epoch 25/100
1/1 [==============================] - 1s 533ms/step - loss: 0.0038 - val_loss: 0.0038
Epoch 26/100
1/1 [==============================] - 1s 524ms/step - loss: 0.0047 - val_loss: 0.0063
Epoch 27/100
1/1 [==============================] - 1s 540ms/step - loss: 0.0057 - val_loss: 0.0069
Epoch 28/100
1/1 [==============================] - 1s 516ms/step - loss: 0.0058 - val_loss: 0.0052
Epoch 29/100
1/1 [==============================] - 1s 528ms/step - loss: 0.0047 - val_loss: 0.0025
Epoch 30/100
1/1 [==============================] - 1s 537ms/step - loss: 0.0031 - val_loss: 0.0013
Epoch 31/100
1/1 [==============================] - 1s 515ms/step - loss: 0.0020 - val_loss: 0.0026
Epoch 32/100
1/1 [==============================] - 1s 520ms/step - loss: 0.0021 - val_loss: 0.0049
Epoch 33/100
1/1 [==============================] - 1s 561ms/step - loss: 0.0029 - val_loss: 0.0054
Epoch 34/100
1/1 [==============================] - 1s 514ms/step - loss: 0.0031 - val_loss: 0.0035
Epoch 35/100
1/1 [==============================] - 1s 541ms/step - loss: 0.0025 - val_loss: 0.0017
Epoch 36/100
1/1 [==============================] - 1s 524ms/step - loss: 0.0017 - val_loss: 0.0016
Epoch 37/100
1/1 [==============================] - 1s 542ms/step - loss: 0.0015 - val_loss: 0.0029
Epoch 38/100
1/1 [==============================] - 1s 538ms/step - loss: 0.0018 - val_loss: 0.0041
Epoch 39/100
1/1 [==============================] - 1s 520ms/step - loss: 0.0019 - val_loss: 0.0041
Epoch 40/100
1/1 [==============================] - 1s 508ms/step - loss: 0.0017 - val_loss: 0.0032
Epoch 41/100
1/1 [==============================] - 1s 535ms/step - loss: 0.0014 - val_loss: 0.0022
Epoch 42/100
1/1 [==============================] - 1s 530ms/step - loss: 0.0013 - val_loss: 0.0016
Epoch 43/100
1/1 [==============================] - 1s 527ms/step - loss: 0.0016 - val_loss: 0.0014
Epoch 44/100
1/1 [==============================] - 1s 525ms/step - loss: 0.0017 - val_loss: 0.0014
Epoch 45/100
1/1 [==============================] - 1s 549ms/step - loss: 0.0015 - val_loss: 0.0014
Epoch 46/100
1/1 [==============================] - 1s 523ms/step - loss: 0.0011 - val_loss: 0.0014
Epoch 47/100
1/1 [==============================] - 1s 532ms/step - loss: 9.4307e-04 - val_loss: 0.0015
Epoch 48/100
1/1 [==============================] - 1s 525ms/step - loss: 0.0010 - val_loss: 0.0015
Epoch 49/100
1/1 [==============================] - 1s 524ms/step - loss: 0.0012 - val_loss: 0.0015
Epoch 50/100
1/1 [==============================] - 1s 570ms/step - loss: 0.0012 - val_loss: 0.0015

<keras.callbacks.History at 0x7f0d848dfe90>

# 予測
predicted = model.predict(g)

plt.figure()
plt.plot(range(25,len(predicted)+25),predicted, color="r", label="predict_data")
plt.plot(range(0, len(f)), f, color="b", label="row_data")
plt.legend()
plt.show()