Skip to content
Snippets Groups Projects
Commit ea55f2c3 authored by Luc Giffon's avatar Luc Giffon
Browse files

update multiview script: polynomial kernel added, new support to transformers

parent 8264e496
Branches
No related tags found
No related merge requests found
Show whitespace changes
Inline Side-by-side
main/experiments/benchmark_vgg_multiview.py
+ 47
− 35
View file @ ea55f2c3
......@@ -2,7 +2,8 @@
Benchmark VGG: Benchmarking deepstrom versus other architectures of the VGG network.
Usage:
benchmark_vgg deepstrom -f name [-q] [--cifar100|--cifar|--mnist|--svhn] [--w-after] [-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]
benchmark_vgg deepstrom -f name [-q] [--cifar100|--cifar|--mnist|--svhn] [--w-after] [-t size] [-d val] [-B nb] [-r] [-a value] [-v size] [-e numepoch] [-s batchsize] [-D reprdim] [-m size]
(-Z|-R|-L|-C|-E|-P|-S|-A|-T|-M) [-g gammavalue] [-c cvalue] [-n] [-y value]
Options:
--help -h Display help and exit.
......@@ -39,33 +40,37 @@ Possible kernels:
-P --chi-square-PD-kernel Says if the Positive definite version of the basic additive chi square kernel should be used for nystrom.
-S --sigmoid-kernel Says it the sigmoid kernel should be used for nystrom.
-A --laplacian-kernel Says if the laplacian kernel should be used for nystrom.
-Z --polynomial-kernel Says if the polynomial kernel should be used for nystrom.
-T --stacked-kernel Says if the kernels laplacian, chi2 and rbf in a stacked setting should be used for nystrom.
-M --sumed-kernel Says if the kernels laplacian, chi2 and rbf in a summed setting should be used for nystrom.
Kernel related:
-g gammavalue --gamma gammavalue The value of gamma for rbf, chi or hyperbolic tangent kernel (deepstrom and deepfriedconvnet)
-c cvalue --intercept-constant cvalue The value of the intercept constant for the hyperbolic tangent kernel.
-y --degree value The value of the degree for polynomial kernel [default: 2]
"""
import logging
import sys
import time as t
import matplotlib.pyplot as plt
import daiquiri
import docopt
import numpy as np
import tensorflow as tf
from sklearn.metrics.pairwise import rbf_kernel, linear_kernel, additive_chi2_kernel, chi2_kernel, laplacian_kernel
from sklearn.metrics.pairwise import rbf_kernel, linear_kernel, additive_chi2_kernel, chi2_kernel, laplacian_kernel, polynomial_kernel
import skluc.data.mldatasets as dataset
from skluc.data.transformation import VGG19Cifar10Transformer, LecunMnistTransformer, VGG19Cifar10BadTransformer, \
VGG19Cifar10BadTransformerV2, VGG19Cifar10BadTransformerV3, VGG19Cifar10BadTransformerV4, VGG19SvhnTransformer, \
VGG19Cifar100Transformer
from skluc.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.tensorflow_.kernel_approximation import nystrom_layer, fastfood_layer
from skluc.tensorflow_.utils import fully_connected, batch_generator, classification_cifar
from skluc.utils import logger, compute_euristic_sigma, compute_euristic_sigma_chi2
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 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, tf_polynomial_kernel
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.utils import logger, compute_euristic_sigma, compute_euristic_sigma_chi2
def print_result():
......@@ -90,7 +95,8 @@ def print_result():
str(DROPOUT),
str(DATASET),
str(WAFTER),
str(FUSING)
str(FUSING),
str(arguments["--degree"])
]
print(",".join(printed_r_list))
exit()
......@@ -151,7 +157,7 @@ def get_gamma_value(arguments, dat, chi2=False):
if __name__ == '__main__':
# todo special treat for each type of execution
logger.debug("Command line: {}".format(' '.join(sys.argv)))
arguments = docopt.docopt(__doc__)
if arguments["--quiet"]:
daiquiri.setup(level=logging.INFO)
......@@ -165,6 +171,7 @@ if __name__ == '__main__':
CHI2_PD_KERNEL = arguments["--chi-square-PD-kernel"]
SIGMOID_KERNEL = arguments["--sigmoid-kernel"]
LAPLACIAN_KERNEL = arguments["--laplacian-kernel"]
POLYNOMIAL_KERNEL = arguments["--polynomial-kernel"]
STACKED_KERNEL = arguments["--stacked-kernel"]
SUMED_KERNEL = arguments["--sumed-kernel"]
VALIDATION_SIZE = int(arguments["--validation-size"])
......@@ -210,39 +217,32 @@ if __name__ == '__main__':
SEED_TRAIN_VALIDATION = SEED
if CIFAR_DATASET:
data = dataset.Cifar10Dataset(validation_size=VALIDATION_SIZE, seed=SEED_TRAIN_VALIDATION)
if BAD_REPR is None or int(BAD_REPR) == 0:
# todo faire quelquechose pour ces "bad repr"
# parametre de bad representation et une seule classe?
transformer = VGG19Cifar10Transformer
elif int(BAD_REPR) == 1:
transformer = VGG19Cifar10BadTransformer
elif int(BAD_REPR) == 2:
transformer = VGG19Cifar10BadTransformerV2
elif int(BAD_REPR) == 3:
transformer = VGG19Cifar10BadTransformerV3
elif int(BAD_REPR) == 4:
transformer = VGG19Cifar10BadTransformerV4
else:
raise ValueError("Not known transformer value: {}".format(BAD_REPR))
transformer = VGG19Transformer(data_name="cifar", cut_layer_name=BAD_REPR)
elif MNIST_DATASET:
data = dataset.MnistDataset(validation_size=VALIDATION_SIZE, seed=SEED_TRAIN_VALIDATION)
transformer = LecunMnistTransformer
# todo rendre conv_pool2 parametrable
trasnformer = 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
transformer = VGG19Transformer(data_name="svhn", cut_layer_name=BAD_REPR)
elif CIFAR100_DATASET:
data = dataset.Cifar100FineDataset(validation_size=VALIDATION_SIZE, seed=SEED_TRAIN_VALIDATION)
transformer = VGG19Cifar100Transformer
transformer = VGG19Transformer(data_name="cifar100", cut_layer_name=BAD_REPR)
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.data_astype(np.float32)
data.labels_astype(np.float32)
data.normalize()
# plt.imshow(data.train.data[0])
# plt.show()
logger.debug("train dataset shape: {}".format(data.train.data.shape))
# exit()
data.apply_transformer(transformer)
data.normalize()
data.to_one_hot()
......@@ -315,12 +315,13 @@ if __name__ == '__main__':
x_slice = tf.expand_dims(x_slice, axis=-1)
logger.debug("x slice shape: {}".format(x_slice.shape))
dataset_slice = data.train.data[:, :, :, i]
dataset_slice = data.train.data[:, :, :, i] # type: np.ndarray
flatttened_dataset_slice = dataset_slice.reshape(dataset_slice.shape[0], -1)
if RBF_KERNEL:
KERNEL = tf_rbf_kernel
KERNEL_NAME = "rbf"
GAMMA = get_gamma_value(arguments, dataset_slice)
GAMMA = get_gamma_value(arguments, flatttened_dataset_slice)
kernel_dict = {"gamma": GAMMA}
elif LINEAR_KERNEL:
KERNEL = tf_linear_kernel
......@@ -347,6 +348,15 @@ if __name__ == '__main__':
KERNEL_NAME = "laplacian"
GAMMA = get_gamma_value(arguments, dataset_slice)
kernel_dict = {"gamma": np.sqrt(GAMMA)}
elif POLYNOMIAL_KERNEL:
KERNEL = tf_polynomial_kernel
KERNEL_NAME = "polynomial2"
kernel_dict = {
"degree": int(arguments["--degree"]),
"coef0": 1,
"gamma": None
} ## beware, those arguments are hard written for the polynomial kernel of scikit learn
elif STACKED_KERNEL:
GAMMA = get_gamma_value(arguments, dataset_slice)
......@@ -410,6 +420,8 @@ if __name__ == '__main__':
elif KERNEL_NAME == "laplacian":
kernel_fct = laplacian_kernel
elif KERNEL_NAME == "polynomial2":
kernel_fct = polynomial_kernel
else:
raise ValueError("Unknown kernel name: {}".format(KERNEL_NAME))
K11 = kernel_fct(nys_subsample_slice, nys_subsample_slice, **kernel_dict)
......@@ -461,7 +473,7 @@ if __name__ == '__main__':
# 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,
train_optimizer = tf.train.AdamOptimizer(learning_rate=1e-3).minimize(cross_entropy,
global_step=global_step)
# calcul de l'accuracy
......@@ -494,7 +506,7 @@ if __name__ == '__main__':
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:
if j % 1 == 0:
logger.info(
"epoch: {}/{}; batch: {}/{}; batch_shape: {}; loss: {}; acc: {}".format(i, NUM_EPOCH, j + 1,
int(data.train[0].shape[
......
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Please register or to comment