def init_conv(use_subset=False):
"Dimensionalities of the filters (number of filters, k_i), d is set from X"
f_1_dim = [20, 5]
f_2_dim = [20, 3]
f_dim = [f_1_dim, f_2_dim]
"The variance to use when initializing filters/weights if not Xavier"
sig_1 = 0.01
sig_2 = 0.01
sig_3 = 0.01
sigmas = [sig_1, sig_2, sig_3]
"Dim of X (d, n_len), and number of classes K"
X_dim = [55, 19]
K = 18
conv = conv_net.conv_net(f_dim, X_dim, K, sigmas)
"Loading the data set"
df = pd.read_pickle('Datasets/X_data.pkl')
char_to_ind = df[0][3]
if (use_subset):
subset_size = 1000
X = np.array(df[0][0][:, :subset_size])
Y = np.array(df[0][1][:, :subset_size])
y = np.array(df[0][2][:subset_size])
else:
X = np.array(df[0][0])
Y = np.array(df[0][1])
y = np.array(df[0][2])
return conv, X, Y, y, char_to_ind
def __init__(self):
# config
self.lr = 1e-4
self.weight_decay = 0.0004
self.batch_size = 50
self.model = conv_net([self.batch_size, 28, 28, 1]) #MNIST input
self.max_epoch = 1
self.param_path = 'param.npy'
def __init__(self, path):
self.model = conv_net([1, 28, 28, 1])
# load_param
param_dict = np.load(path, encoding='bytes').item()
conv1_param = param_dict['conv1']
conv2_param = param_dict['conv2']
fc_param = param_dict['fc']
# fill weights
self.model.conv1.weights = conv1_param['weights']
self.model.conv1.bias = conv1_param['bias']
self.model.conv2.weights = conv2_param['weights']
self.model.conv2.bias = conv2_param['bias']
self.model.fc.weights = fc_param['weights']
self.model.fc.bias = fc_param['bias']
def test(test_file):
image_arr = get_one_image(test_file)
with tf.Graph().as_default():
image = tf.cast(image_arr, tf.float32)
image = tf.image.per_image_standardization(image)
image = tf.reshape(image,[1,32,32,3])
p = conv_net.conv_net(image,1)
logits = tf.nn.softmax(p)
x = tf.placeholder(tf.float32, shape=[32,32,3])
saver = tf.train.Saver()
with tf.Session() as sess:
ckpt = tf.train.get_checkpoint_state('./model/')
if ckpt and ckpt.model_checkpoint_path:
#global_step = ckpt.model_checkpoint_path.split('/')[-1].split[-1]
saver.restore(sess, ckpt.model_checkpoint_path)
print('Loading success')
else:
print('No checkpoint')
prediction =sess.run(logits, feed_dict={x: image_arr})
max_index = np.argmax(prediction)
print(prediction,max_index)
return max_index
import tensorflow as tf import conv_net as cv """ test = tf.keras.datasets.fashion_mnist (train_X, train_Y), (test_X, test_Y) = test.load_data() print(train_X.shape, train_Y.shape) print(train_X[0]) """ test = cv.conv_net(150) test.neural_network_layer.summary()
Gs[2] = G
return Gs
def grad_check(conv):
df = pd.read_pickle('Datasets/X_data.pkl')
X = df[0][0]
Y = df[0][1]
check_grad(conv, X, Y, check_size=10)
f_1_dim = [20, 5]
f_2_dim = [20, 3]
sig_1 = 0.01
sig_2 = 0.01
sig_3 = 0.01
sigmas = [sig_1, sig_2, sig_3]
f_dim = [f_1_dim, f_2_dim]
X_dim = [55, 19]
K = 18
conv = conv.conv_net(f_dim, X_dim, K, sigmas)
grad_check(conv)
config["mini_batch_size"] = 32
config["pickle_file_location"] = config['opath']+'model.zip'
config["output_images_location"] = config['opath'] + 'figs/'
return config
def get_image():
im = imread("boat.jpg")
im = np.swapaxes(im,1,2)
im = np.swapaxes(im,0,1)
im = im[np.newaxis,:,:,:]
print(im.shape)
im = im.astype('float32')
return im
if __name__ == "__main__":
if len(sys.argv) < 2:
print "Arguments needs either be q1 or q2"
exit(0)
config = get_config()
o_conv_net = conv_net(config)
if sys.argv[1] == "q1":
o_conv_net.train()
elif sys.argv[1] == "q2":
o_conv_net.q2(get_image())
else:
print "Arguments can either be q1 or q2"
exit(0)
config["ntest"] = 10000 # max is 10000
config["mini_batch_size"] = 32
config["pickle_file_location"] = config['opath'] + 'model.zip'
config["output_images_location"] = config['opath'] + 'figs/'
return config
def get_image():
im = imread("boat.jpg")
im = np.swapaxes(im, 1, 2)
im = np.swapaxes(im, 0, 1)
im = im[np.newaxis, :, :, :]
print(im.shape)
im = im.astype('float32')
return im
if __name__ == "__main__":
if len(sys.argv) < 2:
print "Arguments needs either be q1 or q2"
exit(0)
config = get_config()
o_conv_net = conv_net(config)
if sys.argv[1] == "q1":
o_conv_net.train()
elif sys.argv[1] == "q2":
o_conv_net.q2(get_image())
else:
print "Arguments can either be q1 or q2"
exit(0)