def check_sigmoid():
X, Y = input_data.loadHandwritingTrainingData(
) #X.shape=(1024,387) Y.shape=(1,387)
layer_types = ['relu', 'tanh', 'sigmoid']
layer_dims = [X.shape[0], 100, 50, Y.shape[0]]
parameters = nn_model.init_params(layer_dims)
gradient_check(X, Y, layer_dims, layer_types, parameters, lambd=1)
def check_dropout():
X, Y, Y_onehot = input_data.loadRandomData()
layer_types = [
'softmax',
]
layer_dims = [X.shape[0], Y_onehot.shape[0]]
parameters = nn_model.init_params(layer_dims)
gradient_check_with_dorpout(X,
Y,
layer_dims,
layer_types,
parameters,
num_params=2)
def check_softmax():
X, Y, Y_onehot = input_data.loadRandomData()
# X.shape=(2, 300), Y.shape=(1,300), Y_onehot=(3,300)
# number of examples = 300, number of classes = 3
layer_types = [
'softmax',
]
layer_dims = [X.shape[0], Y_onehot.shape[0]]
parameters = nn_model.init_params(layer_dims)
gradient_check(X,
Y_onehot,
layer_dims,
layer_types,
parameters,
epsilon=1e-7,
num_params=2,
lambd=1)