import numpy as np
from keras.layers.core import Dense, Dropout, Activation, Flatten
from keras.layers import Convolution1D, MaxPooling1D
from keras.utils import np_utils
from keras.models import Sequential

from sklearn.metrics import mean_squared_error


# Preprocess the data noemalize the data
training_data = np.genfromtxt('employment_data.csv', delimiter=',')
train_data1 = [row[0] for row in training_data]
td = train_data1 / max(train_data1)
X = np.array(td)

#td is of shape(924,)
temp = np.array([td[0:12]])
i = 1
while i < (len(td)-11):
    t_data = np.array([td[i:i+12]])
    temp = np.vstack((temp, t_data))
    i = i+1     
print temp.shape
x_train = temp[:900]
x_test = temp[901:]

# Divide the data in test dataset and train dataset
train_data2 = np.array([row[1] for row in training_data])
train_data3 = train_data2[0:924]
y_train = train_data3[:900]
y_test = train_data3[901:913]


print "This is x_train shape : "
print x_train.shape
print "This is y_train shape : "
print y_train.shape
print "This is x_test shape : "
print x_test.shape
print "This is y_test shape : "
print y_test.shape

# Reshape the data according to neural network
x_train = x_train.reshape((900,12,1))
x_test = x_test.reshape((12,12,1))


# one hot encode outputs categoricaly
y_train = np_utils.to_categorical(y_train)
y_test = np_utils.to_categorical(y_test)
num_classes = y_test.shape[1]

#build the convolutional neural network
print 'Building model...'
model = Sequential()

model.add(Convolution1D(input_dim = 1,
                        nb_filter= 18,
                        filter_length= 3,
                        subsample_length = 1,
                        border_mode='valid',
                        activation='relu',
                        input_length=12))
print model.input_shape

model.add(MaxPooling1D(pool_length=2))
print model.output_shape
model.add(Convolution1D(input_dim = 1,
                        nb_filter= 6,
                        filter_length= 2,
                        subsample_length = 1,
                        border_mode='valid',
                        activation='relu',
                        input_length=12))
print model.output_shape
model.add(Dropout(0.1))
print model.output_shape
model.add(Flatten())
print model.output_shape

model.add(Dense(13, activation='softmax'))
print model.output_shape
model.compile(loss='mse', optimizer='adam', metrics=['accuracy'])

model.fit(x_train, 
          y_train, 
          nb_epoch=300, 
          batch_size = 12, 
          verbose=2, 
          validation_split=0.1)
score = model.evaluate(x_test, y_test, batch_size=12)
print score

predicted = model.predict_classes(x_test)
print predicted