def __init__(self, in_dim, hidden_dim, out_dim, activation, loss_type,
layer_num=0):
NeuralNetwork.__init__(self, activation, loss_type)
args = [self.activation, self.grad_activation]
self.layers = []
self.layers.append(FullyConnectedLayer(in_dim, hidden_dim, *args))
for _ in xrange(layer_num):
self.layers.append(FullyConnectedLayer(hidden_dim, hidden_dim, *args))
if loss_type == 'mse':
self.layers.append(FullyConnectedLayer(hidden_dim, out_dim, *args))
else:
from SoftmaxLayer import SoftmaxLayer
self.layers.append(SoftmaxLayer(hidden_dim, out_dim, *args))
def __init__(self, input, output, output_act, eval_metric):
NeuralNetwork.__init__(self, input, output, output_act, eval_metric)
self.w_size = self.get_weight_vector_length()
self.initialise_cache()
self.W1 = np.random.randn(self.input, self.output) / np.sqrt(
self.input)
self.B1 = np.random.randn(1, self.output) / np.sqrt(
self.output) # bias first layer
self.out = np.zeros((1, self.output)) # output layer for base model
self.final_out = np.zeros(
(1, self.output)) # Final output for the model
def __init__(self):
"""---------------------------------------------------------------------
Desc.: Class Constructor
Args: -
Returns: -
---------------------------------------------------------------------"""
NeuralNetwork.__init__(self, [28 * 28, 10 * 10, 10])
print "----------------------------------------------------------------"
print "Digit Classfier using pytorch nn module"
print "Author: Ankit Manerikar"
print "Written on: 09-21-2017"
print "----------------------------------------------------------------"
print "Loading MNIST Dataset ..."
self.train_images = read_image_file(
'./data/raw/train-images-idx3-ubyte')
self.target_val = read_label_file('./data/raw/train-labels-idx1-ubyte')
print "Dataset Loaded"
print "\nClass initialized"
def __init__(self,
X_data,
Y_data,
n_hidden_neurons=100,
epochs=10,
batch_size=100,
eta=0.1,
lmbd=0.0,
activation_func='relu',
activation_func_out='leaky_relu',
cost_func='MSE',
leaky_a=0.01):
if len(Y_data.shape) == 1:
Y_data = np.expand_dims(Y_data, 1)
n_categories = Y_data.shape[1]
NN.__init__(self, X_data, Y_data, n_hidden_neurons, n_categories,
epochs, batch_size, eta, lmbd, activation_func,
activation_func_out, cost_func)
def __init__(self, input, hidden, output, max_depth, output_act,
eval_metric):
self.hidden = hidden
self.max_depth = max_depth
NeuralNetwork.__init__(self, input, output, output_act, eval_metric)
self.w_size = self.get_weight_vector_length()
self.initialise_cache()
# WEIGHTS FROM INPUT TO FIRST HIDDEN LAYER
self.W1 = np.random.randn(self.input, self.hidden) / np.sqrt(
self.input)
self.B1 = np.random.randn(1, self.hidden) / np.sqrt(self.hidden)
# WEIGHTS FROM LAST HIDDEN LAYER TO OUTPUT LAYER
self.W2 = np.random.randn(self.hidden, self.output) / np.sqrt(
self.hidden)
self.B2 = np.random.randn(1, self.output) / np.sqrt(self.hidden)
self.out = np.zeros((1, self.output)) # output layer for base model
# NOW LETS CREATE ALL OF THE HIDDEN LAYERS
self.h_weights = []
self.h_biases = []
self.h_out = []
for layer in range(self.max_depth):
self.h_weights.append(
np.random.randn(self.hidden, self.hidden) /
np.sqrt(self.hidden))
self.h_biases.append(
np.random.randn(1, self.hidden) / np.sqrt(self.hidden))
self.h_out.append(np.zeros((1, self.hidden)))
self.final_out = np.zeros(
(1, self.output)) # Final output for the model
def __init__(self):
NeuralNetwork.__init__(self, [2, 1])
self.truth_table = [[False, False], [False, True], [True, False],
[True, True]]
self.target_val = [x[0] or x[1] for x in self.truth_table]
def __init__(self):
NeuralNetwork.__init__(self, [1, 1])
self.truth_table = [False, True]
self.target_val = [not x for x in self.truth_table]
def __init__(self, input, output, output_act, eval_metric):
NeuralNetwork.__init__(self, input, output, output_act, eval_metric)
def __init__(self, layers=[]):
NeuralNetwork.__init__(self, layers)
for i in range(self.number_hidden_layers):
self.activation_function[i] = self.ACTIVATION_FUNCTION_SIGMOID
self.output_activation_function = self.ACTIVATION_FUNCTION_SIGMOID
self.loss_function = self.LOSS_FUNCTION_MSE
def __init__(self):
NeuralNetwork.__init__(self)
