diff --git a/main/experiments/benchmark_vgg_tmp.py b/main/experiments/benchmark_vgg_tmp.py
index 89184f0515c64f542e9d3f9eb7e4c86daf24b8ef..a3f60e507214853f483d5fd677b33a1ab70d4c46 100644
--- a/main/experiments/benchmark_vgg_tmp.py
+++ b/main/experiments/benchmark_vgg_tmp.py
@@ -3,7 +3,7 @@ Benchmark VGG: Benchmarking deepstrom versus other architectures of the VGG netw
Usage:
benchmark_vgg dense [-q] [--cifar100|--cifar|--mnist|--svhn] [-f name] [-t size] [-d val] [-B nb] [-a value] [-v size] [-e numepoch] [-s batchsize] [-D reprdim] [-l size]
- benchmark_vgg deepfriedconvnet [-q] [--cifar100|--cifar|--mnist|--svhn] [-f name] [-t size] [-d val] [-B nb] [-a value] [-v size] [-e numepoch] [-s batchsize] [-g gammavalue] [-N nbstack] [-l size]
+ benchmark_vgg deepfriedconvnet [-q] [--cifar100|--cifar|--mnist|--svhn] [-f name] [-t size] [-d val] [-B nb] [-a value] [-v size] [-e numepoch] [-s batchsize] [-g gammavalue] [-N nbstack] [-l size] [-z]
benchmark_vgg deepstrom [-q] [--cifar100|--cifar|--mnist|--svhn] [-f name] [-t size] [-d val] [-B nb] [-r] [-a value] [-v size] [-e numepoch] [-s batchsize] [-D reprdim] [-m size] (-R|-L|-C|-E|-P|-S|-A|-T|-M) [-g gammavalue] [-c cvalue] [-n] [-l size]
Options:
@@ -23,6 +23,7 @@ Dense:
Deepfried convnet:
-N nbstack --nb-stack nbstack The number of fastfood stack for deepfriedconvnet
+ -z --real-fastfood Tell fastfood layer to not update its weights
Deepstrom:
-r --real-nystrom Says if the matrix for deepstrom should be K^(-1/2)
@@ -63,23 +64,24 @@ import numpy as np
import tensorflow as tf
import docopt
from sklearn.metrics.pairwise import rbf_kernel, linear_kernel, additive_chi2_kernel, chi2_kernel, laplacian_kernel
-import skluc.data.mldatasets as dataset
-from skluc.data.transformation import VGG19Cifar10Transformer, LecunMnistTransformer, \
- VGG19SvhnTransformer, VGG19Cifar100Transformer
-from skluc.tensorflow_.kernel_approximation import nystrom_layer, fastfood_layer
-from skluc.tensorflow_.utils import fully_connected, batch_generator, classification_cifar
-from skluc.tensorflow_.kernel import tf_rbf_kernel, tf_linear_kernel, tf_chi_square_CPD, tf_chi_square_CPD_exp, \
+import skluc.main.data.mldatasets as dataset
+from skluc.main.data.transformation.VGG19Transformer import VGG19Transformer
+from skluc.main.data.transformation.LeCunTransformer import LecunTransformer
+from skluc.main.tensorflow_.kernel_approximation.nystrom_layer import nystrom_layer
+from skluc.main.tensorflow_.kernel_approximation.fastfood_layer import fastfood_layer
+from skluc.main.tensorflow_.utils import fully_connected, batch_generator, classification_cifar
+from skluc.main.tensorflow_.kernel import tf_rbf_kernel, tf_linear_kernel, tf_chi_square_CPD, tf_chi_square_CPD_exp, \
tf_chi_square_PD, tf_sigmoid_kernel, tf_laplacian_kernel, tf_stack_of_kernels, tf_sum_of_kernels
-from skluc.utils import logger, compute_euristic_sigma, compute_euristic_sigma_chi2, memory_usage
+from skluc.main.utils import logger, compute_euristic_sigma, compute_euristic_sigma_chi2, memory_usage
-def print_result():
+def print_result(global_acc_val=None, global_acc_test=None, training_time=None, val_eval_time=None, test_eval_time=None):
printed_r_list = [str(NETWORK),
str(global_acc_val),
str(global_acc_test),
str(training_time),
- str(VAL_EVAL_TIME),
- str(TEST_EVAL_TIME),
+ str(val_eval_time),
+ str(test_eval_time),
str(NUM_EPOCH),
str(BATCH_SIZE),
str(OUT_DIM),
@@ -97,7 +99,8 @@ def print_result():
str(CUT_LAYER),
str(TRAIN_SIZE),
str(DROPOUT),
- str(DATASET)
+ str(DATASET),
+ str(REAL_FASTFOOD)
]
print(",".join(printed_r_list))
exit()
@@ -129,10 +132,10 @@ def fct_deepstrom(input_, out_dim, subsample, kernel, kernel_params, w_matrix, n
return out_fc
-def fct_deepfried(input_, nb_stack, sigma):
+def fct_deepfried(input_, nb_stack, sigma, trainable=True):
try:
- return fastfood_layer(input_, sigma, nbr_stack=nb_stack, trainable=True)
+ return fastfood_layer(input_, sigma, nbr_stack=nb_stack, trainable=trainable)
except Exception as e:
logger.critical(e)
print_result()
@@ -153,7 +156,7 @@ def get_gamma_value(arguments, dat, chi2=False):
return gamma_value
-def get_input_classif_deepstrom():
+def get_input_classif_deepstrom(p_x):
logger.info("Selecting {} deepstrom layer function with "
"subsample size = {}, "
"output_dim = {}, "
@@ -190,7 +193,7 @@ def get_input_classif_deepstrom():
added_K11 = np.add(added_K11, rbf_kernel(nys_subsample, nys_subsample, gamma=g_value))
U, S, V = np.linalg.svd(added_K11)
invert_root_K11 = np.dot(U / np.sqrt(S), V).astype(np.float32)
- input_classif = fct_deepstrom(x, OUT_DIM, nys_subsample, KERNEL, kernel_dict,
+ input_classif = fct_deepstrom(p_x, OUT_DIM, nys_subsample, KERNEL, kernel_dict,
w_matrix=invert_root_K11, non_linearity=NON_LINEAR)
elif STACKED_KERNEL:
# here nystrom approximations are stacked
@@ -201,7 +204,7 @@ def get_input_classif_deepstrom():
invert_root_K11 = np.dot(U / np.sqrt(S), V).astype(np.float32)
lst_invert_root_K11.append(invert_root_K11)
stack_K11 = np.vstack(lst_invert_root_K11)
- input_classif = fct_deepstrom(x, OUT_DIM, nys_subsample, KERNEL, kernel_dict,
+ input_classif = fct_deepstrom(p_x, OUT_DIM, nys_subsample, KERNEL, kernel_dict,
w_matrix=stack_K11, non_linearity=NON_LINEAR)
else:
@@ -223,29 +226,144 @@ def get_input_classif_deepstrom():
K11 = kernel_fct(nys_subsample, nys_subsample, **kernel_dict)
U, S, V = np.linalg.svd(K11)
invert_root_K11 = np.dot(U / np.sqrt(S), V).astype(np.float32)
- input_classif = fct_deepstrom(x, OUT_DIM, nys_subsample, KERNEL, kernel_dict, w_matrix=invert_root_K11,
+ input_classif = fct_deepstrom(p_x, OUT_DIM, nys_subsample, KERNEL, kernel_dict, w_matrix=invert_root_K11,
non_linearity=NON_LINEAR)
else:
- input_classif = fct_deepstrom(x, OUT_DIM, nys_subsample, KERNEL, kernel_dict,
+ input_classif = fct_deepstrom(p_x, OUT_DIM, nys_subsample, KERNEL, kernel_dict,
w_matrix=None, non_linearity=NON_LINEAR)
return input_classif
-def get_input_classif_dense():
+def get_input_classif_dense(p_x):
logger.info("Selecting dense layer function with output dim = {} and activation function = {}".format(OUT_DIM, ACTIVATION_FUNCTION))
# two layers is handled outside of here
- input_classif = fct_dense(x, OUT_DIM, two_layers=False, activation_function=ACTIVATION_FUNCTION)
+ input_classif = fct_dense(p_x, OUT_DIM, two_layers=False, activation_function=ACTIVATION_FUNCTION)
return input_classif
-def get_input_classif_deepfriedconvnet():
+def get_input_classif_deepfriedconvnet(p_x):
logger.debug("Selecting deepfriedconvnet layer function")
- input_classif = fct_deepfried(x, NB_STACK, SIGMA)
+ input_classif = fct_deepfried(p_x, NB_STACK, SIGMA, trainable=not REAL_FASTFOOD)
return input_classif
+def main():
+ input_dim, output_dim = data.train[0].shape[1], data.train[1].shape[1]
+
+ x = tf.placeholder(tf.float32, shape=[None, input_dim], name="x")
+ y = tf.placeholder(tf.float32, shape=[None, output_dim], name="label")
+
+ if NETWORK == "dense":
+ input_classif = get_input_classif_dense(x)
+ elif NETWORK == "deepstrom":
+ input_classif = get_input_classif_deepstrom(x)
+ elif NETWORK == "deepfriedconvnet":
+ input_classif = get_input_classif_deepfriedconvnet(x)
+ else:
+ raise Exception("Not recognized network")
+
+ if SIZE_SECOND_LAYER > 0:
+ logger.debug("Add second layer of size: {} and activation {}".format(SIZE_SECOND_LAYER, ACTIVATION_FUNCTION))
+ with tf.variable_scope("second_layer"):
+ input_classif_2nd_layer = fully_connected(input_classif, SIZE_SECOND_LAYER, act=ACTIVATION_FUNCTION,
+ variable_scope="fc")
+ else:
+ logger.debug("No second layer")
+ input_classif_2nd_layer = input_classif
+
+ logger.debug("Add softmax layer for classification")
+ classif, keep_prob = classification_cifar(input_classif_2nd_layer, output_dim)
+
+ # calcul de la loss
+ with tf.name_scope("xent"):
+ cross_entropy = tf.reduce_mean(
+ tf.nn.softmax_cross_entropy_with_logits(labels=y, logits=classif, name="xentropy"),
+ name="xentropy_mean")
+ tf.summary.scalar('loss-xent', cross_entropy)
+
+ # todo learning rate as hyperparameter
+ # calcul du gradient
+ with tf.name_scope("train"):
+ global_step = tf.Variable(0, name="global_step", trainable=False)
+ train_optimizer = tf.train.AdamOptimizer(learning_rate=1e-4).minimize(cross_entropy,
+ global_step=global_step)
+
+ # calcul de l'accuracy
+ with tf.name_scope("accuracy"):
+ predictions = tf.argmax(classif, 1)
+ correct_prediction = tf.equal(predictions, tf.argmax(y, 1))
+ accuracy_op = tf.reduce_mean(tf.cast(correct_prediction, tf.float32))
+ tf.summary.scalar("accuracy", accuracy_op)
+
+ # merged_summary = tf.summary.merge_all()
+
+ init = tf.global_variables_initializer()
+ # Create a session for running Ops on the Graph.
+ # Instantiate a SummaryWriter to output summaries and the Graph.
+ # summary_writer = tf.summary.FileWriter("debug_benchmark_vgg")
+ # Initialize all Variable objects
+ # actual learning
+
+ with tf.Session() as sess:
+ logger.info("Start training")
+ # summary_writer.add_graph(sess.graph)
+ # Initialize all Variable objects
+ sess.run(init)
+ # actual learning
+ # feed_dict_val = {x: data.validation[0], y: data.validation[1], keep_prob: 1.0}
+ global_start = t.time()
+ for i in range(NUM_EPOCH):
+ logger.debug(memory_usage())
+ j = 0
+ start = t.time()
+ for X_batch, Y_batch in batch_generator(data.train[0], data.train[1], BATCH_SIZE, False):
+ feed_dict = {x: X_batch, y: Y_batch, keep_prob: DROPOUT}
+ _, loss, acc = sess.run([train_optimizer, cross_entropy, accuracy_op], feed_dict=feed_dict)
+ if j % 100 == 0:
+ logger.info(
+ "epoch: {}/{}; batch: {}/{}; batch_shape: {}; loss: {}; acc: {}".format(i, NUM_EPOCH, j + 1,
+ int(data.train[0].shape[
+ 0] / BATCH_SIZE) + 1,
+ X_batch.shape, loss,
+ acc))
+ # summary_str = sess.run(merged_summary, feed_dict=feed_dict)
+ # summary_writer.add_summary(summary_str, j)
+ j += 1
+
+ logger.info("Evaluation on validation data")
+ training_time = t.time() - global_start
+ accuracies_val = []
+ i = 0
+ val_eval_start = t.time()
+ for X_batch, Y_batch in batch_generator(data.validation.data, data.validation.labels, 1000, False):
+ accuracy = sess.run([accuracy_op], feed_dict={
+ x: X_batch, y: Y_batch, keep_prob: 1.0})
+ accuracies_val.append(accuracy[0])
+ i += 1
+ global_acc_val = sum(accuracies_val) / i
+ VAL_EVAL_TIME = t.time() - val_eval_start
+
+ logger.info("Evaluation on test data")
+ accuracies_test = []
+ i = 0
+ test_eval_start = t.time()
+ for X_batch, Y_batch in batch_generator(data.test.data, data.test.labels, 1000, False):
+ accuracy = sess.run([accuracy_op], feed_dict={
+ x: X_batch, y: Y_batch, keep_prob: 1.0})
+ accuracies_test.append(accuracy[0])
+ i += 1
+ global_acc_test = sum(accuracies_test) / i
+ TEST_EVAL_TIME = t.time() - test_eval_start
+
+ print_result(global_acc_val=global_acc_val,
+ global_acc_test=global_acc_test,
+ training_time=training_time,
+ val_eval_time=VAL_EVAL_TIME,
+ test_eval_time=TEST_EVAL_TIME)
+
+
if __name__ == '__main__':
- # todo special treat for each type of execution
+ logger.debug("Command line: {}".format(' '.join(sys.argv)))
arguments = docopt.docopt(__doc__)
logger.debug(arguments)
if arguments["--quiet"]:
@@ -276,8 +394,9 @@ if __name__ == '__main__':
CIFAR100_DATASET = bool(arguments["--cifar100"])
MNIST_DATASET = bool(arguments["--mnist"])
SVHN_DATASET = bool(arguments["--svhn"])
- TEST_EVAL_TIME = None
- VAL_EVAL_TIME = None
+ REAL_FASTFOOD = bool(arguments["--real-fastfood"])
+ test_eval_time = None
+ val_eval_time = None
if arguments["--non-linearity"] == "relu":
ACTIVATION_FUNCTION = tf.nn.relu
elif arguments["--non-linearity"] == "tanh":
@@ -312,22 +431,23 @@ if __name__ == '__main__':
SEED_TRAIN_VALIDATION = SEED
if CIFAR_DATASET:
data = dataset.Cifar10Dataset(validation_size=VALIDATION_SIZE, seed=SEED_TRAIN_VALIDATION)
- transformer = VGG19Cifar10Transformer(cut_layer_name=CUT_LAYER)
+ transformer = VGG19Transformer(data_name="cifar10", cut_layer_name=CUT_LAYER)
elif MNIST_DATASET:
data = dataset.MnistDataset(validation_size=VALIDATION_SIZE, seed=SEED_TRAIN_VALIDATION)
- transformer = LecunMnistTransformer()
+ # todo rendre conv_pool2 parametrable
+ transformer = LecunTransformer(data_name="mnist", cut_layer_name="conv_pool_2")
elif SVHN_DATASET:
data = dataset.SVHNDataset(validation_size=VALIDATION_SIZE, seed=SEED_TRAIN_VALIDATION)
- transformer = VGG19SvhnTransformer(cut_layer_name=CUT_LAYER)
+ transformer = VGG19Transformer(data_name="svhn", cut_layer_name=CUT_LAYER)
elif CIFAR100_DATASET:
data = dataset.Cifar100FineDataset(validation_size=VALIDATION_SIZE, seed=SEED_TRAIN_VALIDATION)
- transformer = VGG19Cifar100Transformer(cut_layer_name=CUT_LAYER)
+ transformer = VGG19Transformer(data_name="cifar100", cut_layer_name=CUT_LAYER)
else:
raise ValueError("No dataset specified")
- data.load()
- data.flatten()
- data.to_image() # todo gérer le cas où ce sont déjà des images (les flatteniser dans tous les cas?)
+ data.load() # todo gérer le bug flatten
+ if not data.is_image():
+ data.to_image() # todo gérer le cas où ce sont déjà des images (les flatteniser dans tous les cas?)
data.data_astype(np.float32)
data.labels_astype(np.float32)
data.normalize()
@@ -336,7 +456,8 @@ if __name__ == '__main__':
data.normalize()
data.to_one_hot()
data.flatten()
-
+ data.data_astype(np.float32)
+ data.labels_astype(np.int)
if TRAIN_SIZE is not None:
data.reduce_data_size(int(TRAIN_SIZE))
@@ -417,107 +538,8 @@ if __name__ == '__main__':
else:
raise Exception("Not recognized network")
- input_dim, output_dim = data.train[0].shape[1], data.train[1].shape[1]
-
- x = tf.placeholder(tf.float32, shape=[None, input_dim], name="x")
- y = tf.placeholder(tf.float32, shape=[None, output_dim], name="label")
-
- if NETWORK == "dense":
- input_classif = get_input_classif_dense()
- elif NETWORK == "deepstrom":
- input_classif = get_input_classif_deepstrom()
- elif NETWORK == "deepfriedconvnet":
- input_classif = get_input_classif_deepfriedconvnet()
- else:
- raise Exception("Not recognized network")
-
- if SIZE_SECOND_LAYER > 0:
- logger.debug("Add second layer of size: {} and activation {}".format(SIZE_SECOND_LAYER, ACTIVATION_FUNCTION))
- with tf.variable_scope("second_layer"):
- input_classif_2nd_layer = fully_connected(input_classif, SIZE_SECOND_LAYER, act=ACTIVATION_FUNCTION, variable_scope="fc")
- else:
- logger.debug("No second layer")
- input_classif_2nd_layer = input_classif
-
- logger.debug("Add softmax layer for classification")
- classif, keep_prob = classification_cifar(input_classif_2nd_layer, output_dim)
-
- # calcul de la loss
- with tf.name_scope("xent"):
- cross_entropy = tf.reduce_mean(
- tf.nn.softmax_cross_entropy_with_logits(labels=y, logits=classif, name="xentropy"),
- name="xentropy_mean")
- tf.summary.scalar('loss-xent', cross_entropy)
-
- # todo learning rate as hyperparameter
- # calcul du gradient
- with tf.name_scope("train"):
- global_step = tf.Variable(0, name="global_step", trainable=False)
- train_optimizer = tf.train.AdamOptimizer(learning_rate=1e-4).minimize(cross_entropy,
- global_step=global_step)
-
- # calcul de l'accuracy
- with tf.name_scope("accuracy"):
- predictions = tf.argmax(classif, 1)
- correct_prediction = tf.equal(predictions, tf.argmax(y, 1))
- accuracy_op = tf.reduce_mean(tf.cast(correct_prediction, tf.float32))
- tf.summary.scalar("accuracy", accuracy_op)
-
- merged_summary = tf.summary.merge_all()
-
- init = tf.global_variables_initializer()
- # Create a session for running Ops on the Graph.
- # Instantiate a SummaryWriter to output summaries and the Graph.
- # summary_writer = tf.summary.FileWriter("debug_benchmark_vgg")
- # Initialize all Variable objects
- # actual learning
-
- with tf.Session() as sess:
- logger.info("Start training")
- # summary_writer.add_graph(sess.graph)
- # Initialize all Variable objects
- sess.run(init)
- # actual learning
- # feed_dict_val = {x: data.validation[0], y: data.validation[1], keep_prob: 1.0}
- global_start = t.time()
- for i in range(NUM_EPOCH):
- logger.debug(memory_usage())
- j = 0
- start = t.time()
- for X_batch, Y_batch in batch_generator(data.train[0], data.train[1], BATCH_SIZE, False):
- feed_dict = {x: X_batch, y: Y_batch, keep_prob: DROPOUT}
- _, loss, acc = sess.run([train_optimizer, cross_entropy, accuracy_op], feed_dict=feed_dict)
- if j % 100 == 0:
- logger.info("epoch: {}/{}; batch: {}/{}; batch_shape: {}; loss: {}; acc: {}".format(i, NUM_EPOCH, j+1, int(data.train[0].shape[0]/BATCH_SIZE)+1, X_batch.shape, loss, acc))
- # summary_str = sess.run(merged_summary, feed_dict=feed_dict)
- # summary_writer.add_summary(summary_str, j)
- j += 1
-
- logger.info("Evaluation on validation data")
- training_time = t.time() - global_start
- accuracies_val = []
- i = 0
- val_eval_start = t.time()
- for X_batch, Y_batch in batch_generator(data.validation.data, data.validation.labels, 1000, False):
- accuracy = sess.run([accuracy_op], feed_dict={
- x: X_batch, y: Y_batch, keep_prob: 1.0})
- accuracies_val.append(accuracy[0])
- i += 1
- global_acc_val = sum(accuracies_val) / i
- VAL_EVAL_TIME = t.time() - val_eval_start
-
- logger.info("Evaluation on test data")
- accuracies_test = []
- i = 0
- test_eval_start = t.time()
- for X_batch, Y_batch in batch_generator(data.test.data, data.test.labels, 1000, False):
- accuracy = sess.run([accuracy_op], feed_dict={
- x: X_batch, y: Y_batch, keep_prob: 1.0})
- accuracies_test.append(accuracy[0])
- i += 1
- global_acc_test = sum(accuracies_test) / i
- TEST_EVAL_TIME = t.time() - test_eval_start
-
-
- print_result()
-
+ try:
+ main()
+ except Exception as e:
+ print_result()
+ raise e
\ No newline at end of file