def testA():
#this part shows the abiliy of autoencoders to recognise piecewise patterns
x, y = dh.createSetFromCSV('dataset//ac1.csv')
xt, yt = dh.createSetFromCSV('dataset//ac2.csv')
x = np.array(x, np.float32)
ae = Autoencoder(sizes=[30, 20], lr=0.1)
ae.train(x, 1000, output=True)
print("Showing results for original set:")
ae.use(x)
print("Showing results for test set:")
ae.use(xt)
return h
def fprop(self, input_d):
for layer in range(len(self.weights)):
input_d = self.model(input_d, layer)
return input_d
def use(self, useset):
predict_op = tf.argmax(self.fprop(useset), 1)
return self.sess.run(predict_op)
def accuracy(self, testset, y):
correct_prediction = tf.equal(tf.argmax(y, 1), self.use(testset))
return self.sess.run(
tf.reduce_mean(tf.cast(correct_prediction, tf.float32)))
if __name__ == "__main__":
x, y = dh.createSetFromCSV('training.csv')
testset, y_ = dh.createSetFromCSV('test.csv')
ann = NeuralNetwork(sizes=[500], lr=0.01)
ann.train(x,
y,
1500,
printInterval=100,
output=True,
save=False,
load=False)
# print("Accuracy:",ann.accuracy(testset,y_))
# dh.csvOutput(ann.use(testset),'output//output_ann.csv')
self.Y: Y_labels
}))
return
def model(self, input, i):
h = tf.nn.sigmoid(tf.matmul(input, self.weights[i]) + self.biases[i])
return h
def fprop(self, input_d):
for layer in range(len(self.weights)):
input_d = self.model(input_d, layer)
return input_d
def use(self, useset):
predict_op = tf.argmax(self.fprop(useset), 1)
return self.sess.run(predict_op)
def accuracy(self, testset, y):
correct_prediction = tf.equal(tf.argmax(y, 1), self.use(testset))
return self.sess.run(
tf.reduce_mean(tf.cast(correct_prediction, tf.float32)))
x, y = dh.createSetFromCSV('dataset//training.csv')
testset, y_ = dh.createSetFromCSV('dataset//test.csv')
ann = NeuralNetwork(sizes=[3], lr=0.2)
ann.train(x, y, 1000, output=False)
print("Accuracy:", ann.accuracy(testset, y_))
dh.csvOutput(ann.use(testset), 'output//output_ann.csv')