fully connected example - MNIST.py

#%% load dataset

import numpy as np
import DL

# change the directory
Label_Train, Features_Train, Label_Test, Features_Test = DL.ReadFile("H:\\4th comp\\NN\\MNISTcsv")

# features dimensions (m, c, h, w)


#%% training
batch_size = 64
num_epochs = 10
num_classes = 10
hidden_units = 300

input_dimensions = (28, 28, 1)

# change each label from scaler value to vector( 2 ---> [0, 0, 1, 0, 0, ...] ) (hot one)
Label_Train_hotone = DL.hot_one(Label_Train, num_classes)

model = DL.model()
model.input_dims(input_dimensions)
model.add('flatten')
model.add('Relu', hidden_units)
model.add('Linear', num_classes)
optim = DL.optimizer('gd',0.5,0.5)
# optim = DL.optimizer('adam',0.001)
loss_fn = DL.loss_Function('SoftmaxCrossEntropy')
loss_fn.setLambda(0)

model.fit(Features_Train, Label_Train_hotone,
          batch_size, num_epochs, optim, loss_fn)


#%% testing

# test on the same trained data set
predicted_labels = np.argmax(model.predict(Features_Train), axis=0)
accuracy = DL.accuracy(predicted_labels, Label_Train)
print("Accuracy of training dataset = {:.2f}%".format(accuracy*100))

# test on the test data set
predicted_labels = np.argmax(model.predict(Features_Test), axis=0)
accuracy = DL.accuracy(predicted_labels, Label_Test)
print("Model Accuracy = {:.2f}%".format(accuracy*100))


#%% store and load model

# DL.store(model, "FC MNIST model") # store

# model = DL.load("FC MNIST model") # load