def build_CompressModel():
print 'STEP 1 start...'
dataset = Nikkei(dataset_type=params['experiment_type'],
brandcode=params['STEP3']['brandcode'])
# pdb.set_trace()
index = T.lscalar() # index to a [mini]batch
x = T.matrix('x') # the data is presented as rasterized images
if params['STEP1']['model'] == 'rbm':
model = RBM(input=x,
n_visible=dataset.phase1_input_size,
n_hidden=params['STEP1']['n_hidden'],
reg_weight=params['STEP1']['beta'])
train_rbm(input=x,
model=model,
dataset=dataset,
learning_rate=params['STEP1']['learning_rate'],
outdir=model_dirs['STEP1'])
else:
model = SparseAutoencoder(input=x,
n_visible=dataset.phase1_input_size,
n_hidden=params['STEP1']['n_hidden'],
beta=params['STEP1']['beta'])
train_sae(input=x,
model=model,
dataset=dataset,
learning_rate=params['STEP1']['learning_rate'],
outdir=model_dirs['STEP1'])
def load_model(model_type='sae', input=None, params_dir=None):
params = cPickle.load(open(params_dir))
if model_type == 'rbm':
model = RBM(input=input, params=params)
else:
model = SparseAutoencoder(input=input, params=params)
return model
def load_model(input=None, params_dir=None):
params = cPickle.load(open(params_dir))
model = SparseAutoencoder(input=input, params=params)
return model
def build_CompressModel():
print 'STEP 1 start...'
dataset = Nikkei(dataset_type=params['dataset_type'],
brandcode=params['STEP3']['brandcode'])
# pdb.set_trace()
index = T.lscalar() # index to a [mini]batch
x = T.matrix('x') # the data is presented as rasterized images
if params['STEP1']['model'] == 'rbm':
model = RBM(input=x,
n_visible=dataset.phase1_input_size,
n_hidden=params['STEP1']['n_hidden'],
reg_weight=params['STEP1']['reg_weight'],
corruption_level=params['STEP1']['corruption_level'])
train_rbm(input=x,
model=model,
dataset=dataset,
learning_rate=params['STEP1']['learning_rate'],
outdir=model_dirs['STEP1'],
batch_size=params['STEP1']['batch_size'])
elif params['STEP1']['model'] == 'sda':
sda_params = {
'dataset':
dataset,
'hidden_layers_sizes':
[params['STEP1']['n_hidden'], params['STEP1']['n_hidden'] / 2],
'pretrain_lr':
params['STEP1']['learning_rate'],
'pretrain_batch_size':
params['STEP1']['batch_size'],
'pretrain_epochs':
5,
'corruption_levels': [0.5, 0.5],
'k':
None,
'y_type':
0,
'sparse_weight':
params['STEP1']['reg_weight']
}
model = SdA.compress(sda_params)
pre_params = get_model_params(model)
while (True):
try:
f_out = open(model_dirs['STEP1'], 'w')
f_out.write(cPickle.dumps(model, 1))
f_out.close()
break
except:
pdb.set_trace()
else:
model = SparseAutoencoder(
input=x,
n_visible=dataset.phase1_input_size,
n_hidden=params['STEP1']['n_hidden'],
reg_weight=params['STEP1']['reg_weight'],
corruption_level=params['STEP1']['corruption_level'])
train_sae(input=x,
model=model,
dataset=dataset,
learning_rate=params['STEP1']['learning_rate'],
outdir=model_dirs['STEP1'],
batch_size=params['STEP1']['batch_size'])