モジュールをインポート
import numpy as np import matplotlib.pyplot as plt from keras.models import Sequential from keras.layers import Dense, Dropout from keras.layers.convolutional import Conv1D, UpSampling1D from keras.layers.pooling import MaxPooling1D
サンプルデータ生成
LEN_DATA=10000 timeline = np.arange(LEN_DATA) raw_data = np.sin(timeline * 1 / 100) + np.cos(timeline * 3 / 100)/2 raw_data = raw_data + (np.random.rand(len(timeline)) * 0.3)# ノイズ項 plt.style.use(['default']) plt.plot(timeline[:5000], raw_data[:5000]) plt.xlabel("time") plt.ylabel("data") plt.show()
入力データに変換
LEN_INPUT=64 LEN_OUTPUT=16 LEN_TEST=5000 DATA_MAX=1.8 DATA_MIN=-1.5 input_data = [] output_data = [] raw_data=(raw_data-DATA_MIN)/(DATA_MAX-DATA_MIN) raw_data=np.clip(raw_data,0.0,1.0) for n in range(LEN_DATA-LEN_INPUT-LEN_OUTPUT): input_data.append(raw_data[n:n+LEN_INPUT]) output_data.append(raw_data[n+LEN_INPUT:n+LEN_INPUT+LEN_OUTPUT]) input_data = np.array(input_data[:LEN_TEST]) output_data = np.array(output_data[:LEN_TEST]) print(input_data.shape) print(output_data.shape) train_X = np.reshape(input_data, (-1, LEN_INPUT, 1)) train_Y = np.reshape(output_data, (-1, LEN_OUTPUT, 1)) print(train_X.shape) print(train_Y.shape)
NNモデル定義
model = Sequential() model.add(Conv1D(64, 8, padding='same', input_shape=(64, 1), activation='relu')) model.add(MaxPooling1D(2, padding='same')) model.add(Conv1D(64, 8, padding='same', activation='relu')) model.add(MaxPooling1D(2, padding='same')) model.add(Conv1D(32, 8, padding='same', activation='relu')) model.add(Conv1D(1, 8, padding='same', activation='tanh')) model.compile(loss='mse', optimizer='adam') model.summary()
学習
EPOCHS=100 #学習 history = model.fit(train_X, train_Y, validation_split=0.1, epochs=EPOCHS) #損失プロット plt.plot(range(EPOCHS), history.history['loss'], label='loss') plt.plot(range(EPOCHS), history.history['val_loss'], label='val_loss') plt.xlabel('epoch') plt.ylabel('loss') plt.legend() plt.show()
推論テスト
start = LEN_TEST sheed = np.reshape(raw_data[start:start+LEN_INPUT], (1, LEN_INPUT, 1)) prediction = sheed sheed2=sheed prediction2 = sheed2 for i in range(20): res = model.predict(sheed) sheed = np.concatenate((sheed[:, 16:, :], res), axis=1) prediction = np.concatenate((prediction, res), axis=1) res2 = model.predict(sheed2) sheed2 = np.reshape(raw_data[start+(i+1)*16:start+64+(i+1)*16], (1, 64, 1)) prediction2 = np.concatenate((prediction2, res2), axis=1) print(prediction.shape) print(prediction2.shape) predictor = np.reshape(prediction, (-1)) predictor2 = np.reshape(prediction2, (-1)) print(predictor.shape) print(predictor2.shape) plt.plot(range(len(predictor)), raw_data[start:start + len(predictor)], label='real') plt.plot(range(len(predictor)), predictor, label='predict') plt.plot(range(len(predictor2)), predictor2, label='predict2') plt.legend() plt.show()
