hidden_nodes_list = [400] # default 100
learning_rate_list = [0.05]
best_performance = 0
best_learning_rate = 0
best_hidden_nodes = 0
best_nr_of_epochs = 0
timestamp = time.time()
for h_nodes in hidden_nodes_list:
print("hidden nodes: ", h_nodes)
for lr in learning_rate_list:
print(" learning rate: ", lr)
mnist = Mnist(h_nodes, lr)
performance_list = []
for e in range(epochs):
print(" train epoch: ", e + 1)
# train the neural network
train_start_time = time.time()
mnist.train()
print(" training took:", time.time() - train_start_time, "ms")
# test the neural network
test_start_time = time.time()
performance = mnist.test()
print(" test took:", time.time() - test_start_time, "ms")
Since Hopfield networks are not supervised models, we must
turn classification into a memory recall task. To do this,
we feed the network augmented vectors containing both the
image and a one-hot vector representing the class.
"""
# Import libraries
import numpy as np
import os
import sys
import urllib
import time
import tensorflow as tf
import FileProcess as fipr
from Mnist import Mnist
mnist = Mnist()
from update import hebbian_update
from update import extended_storkey_update
from network import Network
# from tensorflow.examples.tutorials.mnist import input_data
BATCH_SIZE = 100
# Start counting time
start_time = time.clock()
# open and load csv files
time_load_start = time.clock()
X_train, y_train = fipr.load_csv("train_file.csv", True)
X_test, y_test = fipr.load_csv("test_file.csv", True)
from Mnist import Mnist
from Network import Network
m = Mnist('train-images-idx3-ubyte', 'train-labels-idx1-ubyte',
't10k-images-idx3-ubyte', 't10k-labels-idx1-ubyte')
net = Network([784, 30, 10])
net.SGD(m.train_data, 30, 10, 3.0, test_data=m.test_data)
### Build CNN for Data with New Patterns ###
# Import libraries
import numpy as np
import os
import sys
import urllib
import time
import tensorflow as tf
import FileProcess as fipr
from Mnist import Mnist
mnist = Mnist()
sess = tf.InteractiveSession()
# Start counting time
start_time = time.clock()
# open and load csv files
time_load_start = time.clock()
X_train, y_train = fipr.load_csv("train_file.csv", True)
X_test, y_test = fipr.load_csv("test_file.csv", True)
#y_train = y_train.flatten()
#y_test = y_test.flatten()
time_load_end = time.clock()
print("Loading finished, loading time: %g seconds" %
(time_load_end - time_load_start))
X_test_even, y_test_even = fipr.load_csv("test_file_even.csv", True)
from Mnist import Mnist
from NN import NN
import numpy as np
dataset = Mnist("train-images.idx3-ubyte", "train-labels.idx1-ubyte")
X_train = dataset.get_images((0, 20000))
y_train = dataset.get_labels((0, 20000))
X_val = dataset.get_images((20001, 21001))
y_val = dataset.get_labels((20001, 21001))
network = NN()
network.train(X_train, y_train, iterations=10000, batch_size=1000)
print(np.mean(network.predict(X_val) == y_val))
### Build CNN for Experiments ###
# Import libraries
import numpy as np
import os
import sys
import urllib
import time
import tensorflow as tf
sess = tf.InteractiveSession()
from Mnist import Mnist
mnist = Mnist()
training_data, training_labels = mnist.load_training_batch()
# Uniformed Distribution of Data
training_data, training_labels = mnist.balance_data(training_data,
training_labels)
print(type(training_labels))
print(type(training_labels[0, 0]))
print(training_labels.shape)
mnist.print_sample_distribution(training_labels)
training_labels = mnist.make_one_hot(training_labels)
#print("a sample of training label:" % training_labels)
test_data, test_labels = mnist.load_test_batch()
test_data = np.reshape(test_data, (-1, 784))
test_labels = mnist.make_one_hot(test_labels)
def start():
m = Mnist('train-images-idx3-ubyte', 'train-labels-idx1-ubyte',
't10k-images-idx3-ubyte', 't10k-labels-idx1-ubyte')
net = Network([784, 30, 10])
net.SGD(m.train_data, 30, 10, 3.0, test_data=m.test_data)
# fileName: mnsit
# author: xiaofu.qin
# create at 2017/12/17
# description:
import keras
from keras.models import Sequential
from keras.layers import Dense, Activation
from keras.optimizers import SGD
from keras.losses import categorical_crossentropy
from Mnist import Mnist
train_images = Mnist.extract_images("../mnist-data/train-images.idx3-ubyte")
train_labels = Mnist.extract_labels("../mnist-data/train-labels.idx1-ubyte")
train_labels = keras.utils.to_categorical(train_labels, num_classes=10)
test_images = Mnist.extract_images("../mnist-data/t10k-images.idx3-ubyte")
test_labels = Mnist.extract_labels("../mnist-data/t10k-labels.idx1-ubyte")
test_labels = keras.utils.to_categorical(test_labels, num_classes=10)
print("extracted images and labels")
# define the model
model = Sequential()
# add one dense layer
model.add(Dense(100, activation="relu", input_dim=784))
model.add(Dense(500, activation="relu"))
model.add(Dense(10, activation="softmax"))
# sgd = SGD(lr=0.02, momentum=0.95)