LSTM variation 2

python/lstm_nn.py

@@ -0,0 +1,54 @@
+import numpy as np
+import pandas as pd
+import h5py
+from keras.models import load_model
+from keras.models import Sequential
+from keras.layers import Dense
+from keras.utils import np_utils
+from keras.layers import LSTM
+from sklearn.preprocessing import LabelEncoder
+from scipy.stats import mode
+
+time_step = 8
+
+def generateDataFormat1(X):
+	X = X.values
+	print X.shape
+	#dataX = np.zeros(shape=(X.shape[0]-time_step-1,time_step,X.shape[1]))
+	dataX = np.zeros(shape=(10520,8,40))
+	print dataX.shape
+	for i in range(0,dataX.shape[0]):
+		dataX[i] = X[i:i+time_step]
+	return dataX
+
+def generateLabelFormat1(Y):
+	print Y.shape
+	dataY = Y[8:]
+	print dataY.shape
+	return dataY
+
+dataframe = pd.read_csv("../data/training/pca_data_v2/pca_subject1.csv")
+#dataframe = dataframe.sample(frac=1).reset_index(drop=True)
+X = dataframe.iloc[:,:-1]
+Y = dataframe.iloc[:,-1]
+encoder = LabelEncoder()
+encoder.fit(Y)
+encoded_Y = encoder.transform(Y)
+dummy_y = np_utils.to_categorical(encoded_Y)
+print dummy_y
+dataX = generateDataFormat1(X)	#8 data instances for one data classification
+dataY = generateLabelFormat1(dummy_y) # encoded labels starting 8th sample
+# now need to shuffle dataX and dataY with the same shuffling vector.
+# generate random integers of dataX.shape[0] and store in a list variable and use as index to shuffle data with windows
+shuffle_vector = np.random.randint(low=0, high=dataX.shape[0], size=dataX.shape[0])
+dataX=dataX[shuffle_vector]
+dataY=dataY[shuffle_vector]
+# now design the LSTM model and while fiting set shuffle=False
+model = Sequential()
+model.add(LSTM(32,input_shape=(dataX.shape[1],dataX.shape[2])))
+model.add(Dense(dataY.shape[1], activation='sigmoid'))
+model.compile(loss='categorical_crossentropy', optimizer='adam', metrics=['accuracy'])
+model.fit(dataX, dataY, nb_epoch=50, batch_size=10, verbose=2, shuffle=False)
+# save the trained model
+path = '../trained_models/feedForwardSubject1_LSTM2.h5'
+model.save(path)