import numpy as np from sklearn.model_selection import train_test_split from tensorflow.keras.models import Sequential, Model from tensorflow.keras.layers import Conv2D, MaxPooling2D, Dropout, LSTM, Flatten, Dense

Step 1: Input data

data = np.loadtxt('data.txt', delimiter=',') X = data[:, :23] Y = data[:, -1]

Step 2: Define hyper-parameters

optimizer = 'adam' rate = 0.2 feature_layers = 64 poolsize = (2, 2) batchsize = 32 epochs = 10 units = 128 activation = 'relu' recurrent_activation = 'sigmoid' lossfunction = 'categorical_crossentropy'

Step 3: Initialize model

model = Sequential()

Step 4: Normalization

X = (X - np.mean(X)) / np.std(X)

Step 5: Convolution layer

model.add(Conv2D(feature_layers, kernel_size=(3, 3), activation='relu', input_shape=(23, 1, 1))) model.add(MaxPooling2D(pool_size=poolsize)) model.add(Dropout(rate))

Step 6: Compile and train Convolution layer

model.compile(optimizer=optimizer, loss='categorical_crossentropy', metrics=['accuracy']) model.fit(X, Y, epochs=epochs, batch_size=batchsize)

Step 7: Extract features from Convolution layer

feature_model = Model(inputs=model.input, outputs=model.get_layer('conv2d').output) features = feature_model.predict(X)

Step 8: Build BiGRU model

bigru_model = Sequential() bigru_model.add(LSTM(units=units, activation=activation, recurrent_activation=recurrent_activation, input_shape=(features.shape[1], features.shape[2]))) bigru_model.add(Flatten()) bigru_model.add(Dense(8, activation='softmax'))

Step 9: Compile and train BiGRU model

bigru_model.compile(loss=lossfunction, optimizer=optimizer) bigru_model.fit(features, Y, batch_size=batchsize, epochs=epochs)

Step 10: Evaluate the model

test_loss, test_acc = bigru_model.evaluate(features, Y) print('Test accuracy:', test_acc)