LAYER_SIZE = [256, 512, 1024]
N_LAYERS = [2, 3, 4]
ITERATIONS = 5
datasets = load_data(n_frames=N_FRAMES)
for _ in range(ITERATIONS):
for n_layers in N_LAYERS:
for layer_size in LAYER_SIZE:
test_score, val_score = test_GRBM_DBN(
finetune_lr=0.1,
pretraining_epochs=[225, 75],
pretrain_lr=[0.002, 0.02],
k=1,
weight_decay=0.0002,
momentum=0.9,
batch_size=128,
datasets=datasets,
hidden_layers_sizes=n_layers * [layer_size],
load=False,
n_ins=39 * N_FRAMES,
n_outs=120,
filename=('../data/speech_%d_%d_%d.pickle' %
(N_FRAMES, layer_size, n_layers)))
log = '../data/speech.log'
with open(log, 'a') as f:
f.write(
'N_FRAMES=%d, LAYER_SIZE=%d, n_layers=%d, test_score=%f%%, val_score=%f%%\n'
% (N_FRAMES, layer_size, n_layers, test_score, val_score))
# -*- coding: utf-8 -*-
from GRBM_DBN import test_GRBM_DBN
from GRBM_DBN import GRBM_DBN
from load_data_MNIST import load_data
from load_data_MNIST import load_raw_data
datasets = load_data()
#
# UCZYMY SIEĆ
#
test_score, val_score = test_GRBM_DBN(finetune_lr=0.1, pretraining_epochs=[1, 1], pretrain_lr=[0.002, 0.02], k=1, weight_decay=0.0002,
momentum=0.9, batch_size=128, datasets=datasets, hidden_layers_sizes=[784,784], finetune = False,
saveToDir = '../results/MNIST/', loadModelFromFile = '', verbose = True)
#
# UŻYCIE WYUCZONEJ SIECI
#
dbn = GRBM_DBN.load('../results/MNIST/pretrained_model')
train_set, valid_set, test_set = load_raw_data()
#klasyfikacja pierwszych 13 wzorców
print dbn.classify(train_set[0][1:13])
#realne klasy pierwszych 13 wzorców
print train_set[1][1:13]
from GRBM_DBN import test_GRBM_DBN
from load_data_MNIST import load_data
LAYER_SIZE = [256, 512, 1024]
N_LAYERS = [2, 3, 4]
ITERATIONS = 5
datasets = load_data()
for _ in range(ITERATIONS):
for n_layers in N_LAYERS:
for layer_size in LAYER_SIZE:
test_score, val_score = test_GRBM_DBN(finetune_lr=0.1, pretraining_epochs=[225, 75],
pretrain_lr=[0.002, 0.02], k=1, weight_decay=0.0002,
momentum=0.9, batch_size=128, datasets=datasets,
hidden_layers_sizes=n_layers*[layer_size], load=False,
filename=('../data/MNIST_%d_%d.pickle'%(layer_size, n_layers)))
log = '../data/MNIST.log'
with open(log, 'a') as f:
f.write('LAYER_SIZE=%d, n_layers=%d, test_score=%f%%, val_score=%f%%\n' % (layer_size, n_layers, test_score, val_score))
from GRBM_DBN import test_GRBM_DBN
from load_data_MNIST import load_data
datasets = load_data()
test_GRBM_DBN(finetune_lr=0.2, pretraining_epochs=[70, 40],
pretrain_lr=[0.0002, 0.002], k=1, weight_decay=0.02,
momentum=0.8, batch_size=20, datasets=datasets,
hidden_layers_sizes=[784,784], load=False, save=True,
filename='../data/MNIST/GRBM200/layer-1.pickle',
finetune=True, pretraining_start=0, pretraining_stop=2, verbose=True)
from GRBM_DBN import test_GRBM_DBN
from load_data_MNIST import load_data
datasets = load_data()
test_GRBM_DBN(finetune_lr=0.2,
pretraining_epochs=[200, 50],
pretrain_lr=[0.0001, 0.002],
k=1,
weight_decay=0.002,
momentum=0.7,
batch_size=20,
datasets=datasets,
hidden_layers_sizes=[784, 784],
load=False,
save=True,
filename='../data/MNIST/GRBM200/layer-4.pickle',
finetune=True,
pretraining_start=0,
pretraining_stop=2,
verbose=True)
