MLP Numpy Module

class mlp.MLP(seed=99, verbose=False, restore=False)

Bases: object

Initialize MLP object

Parameters:

• seed (int) – seed for random
• verbose (bool) – If want to print help message during using MLP object
• restore (bool) – Restore from checkpoint or not?

nn_architecture = [{'input_dim': 1296, 'output_dim': 100, 'activation': 'relu'}, {'input_dim': 100, 'output_dim': 10, 'activation': 'softmax'}]

static relu(z)

Function which apply relu function to current z values

Parameters:z (np.array) – current value Returns:current value after activation function a_curr Return type:np.array

static relu_backward(d_a, z)

Function which apply relu function to current error value for backpropagation process

Parameters:

• d_a (np.array) – current error value
• z (np.array) – current value before activation function

Returns:

error before activation function

Return type:

np.array

static softmax(z)

Function which apply softmax function to current z values

Parameters:z (np.array) – current value Returns:current value after activation function a_curr Return type:np.array

softmax_backward(d_a, z)

Softmax backward propagation

Parameters:

• d_a (np.array) – current error value
• z (np.array) – current value before activation function

Returns:

error before activation function

Return type:

np.array

single_layer_forward_propagation(a_prev, w_curr, b_curr, activation='relu')

Function for make forward propagation through one layer in Neural Network

Parameters:

• a_prev (np.array) – previous value after activation function (input)
• w_curr (np.array) – current weights value
• b_curr (np.array) – current bias value
• activation (str) – type of activation function

Returns:

a_curr and value before activation function z_curr

Return type:

tuple

full_forward_propagation(x)

Function for make full forward propagation through Neural Network

Parameters:x (np.array) – training data Returns:None Return type:None

get_cost_value(targets, epsilon=1e-10) → float

Cross-entropy loss function

Parameters:

• targets (np.array) – ground truth
• epsilon (float) – hiperparater

Returns:

cross_entropy value

Return type:

float

get_accuracy_value(targets: numpy.array) → float

Function to count accuracy value

Parameters:targets (np.array) – ground truth Returns:accuracy value Return type:float

single_layer_backward_propagation(d_a_curr, w_curr, z_curr, a_prev, activation='relu')

Function for make single layer backward propagation

Parameters:

• d_a_curr (np.array) – current error value
• w_curr (np.array) – current weights value
• z_curr (np.array) – current value before activation function
• a_prev (np.array) – previous neuron autput value
• activation (str) – type of activation function

Returns:

d_a_prev, d_w_curr, db_curr

Return type:

tuple

full_backward_propagation(targets: numpy.array) → dict

Function for make full backward propagation through Neural Network

Parameters:targets (np.array) – ground truth Returns:gradients values to update weights of NN Return type:dict

update(grads_values, learning_rate)

Update NN weights and bias

Parameters:

• grads_values (np.array) – gradient value
• learning_rate (float) – current learning_rate

train(train_data: numpy.array, train_labels: numpy.array, epochs: int = 1000, learning_rate: float = 0.001) → tuple

Function which trains model

Parameters:

• train_data (np.array) – x data for training
• train_labels (np.array) – y data for training
• epochs (int) – number of epochs for training
• learning_rate (float) – hiperparameter of training

Returns:

(self.params_values, self.cost_history, self.accuracy_history)

Return type:

tuple

one_hot_encode(labels: numpy.array)

Function which convert labels to vector

Parameters:labels (np.array) – np.array of labels to convert Returns:converted labels in one_hot_encode Return type:np.array

one_hot_decode(labels) → numpy.array

Function which convert vector to labels

Parameters:labels (np.array) – np.array of predicted labels to convert Returns:converted labels from one_hot_encode Return type:np.array

test(x_input, ground_truth)

Function for test model

Parameters:

• x_input (np.array) – test data
• ground_truth (np.array) – test labels

Returns:

test accuracy

Return type:

float

predict(x_input)

Funtion to predict output for given input

Parameters:x_input (np.array) – image data Returns:predition vector Return type:np.array

save()

Funtion for saving model current variables to files

load()

Funtion for loading model current variables to files

Return Home