In [ ]:

    

import pickle

import matplotlib.pyplot as plt
import numpy as np
import sys
sys.path.append("../") # go to parent dir
from autoencoder.mlp import Network

test_data = np.load('test.npy')
train_data = np.load('train.npy')


    

In [ ]:

    

batch_size = 32
learning_rate = 0.2
l2 = 0.0
epochs = 5
hidden=100

results = []
for size in [1,2,4,5,8,10,16,20,32,40,60]:
    print("Size = {}".format(size * 1000))
    mlp = Network(100)
    mlp.drop = 0
    train = train_data[:, :size * 1000]
    mlp.train(train, learning_rate, epochs, batch_size, l2, test_data)
    results.append(mlp)
    
    mlp.hidden_weights = np.random.randn(hidden, 784) / np.sqrt(784)
    mlp.output_weights = np.random.randn(784, hidden) / np.sqrt(hidden)

    mlp.hidden_bias = np.ones(shape=(hidden, 1))
    mlp.output_bias = np.ones(shape=(784, 1))


    

In [ ]:

    




    

In [ ]: