def load(dset):
if dset in ['jsb', 'nottingham', 'musedata', 'piano']:
musicdata = loadDataset(dset)
dataset = {}
for k in ['train', 'valid', 'test']:
dataset[k] = {}
dataset[k]['tensor'] = musicdata[k]
dataset[k]['mask'] = musicdata['mask_' + k]
dataset['data_type'] = musicdata['data_type']
dataset['dim_observations'] = musicdata['dim_observations']
elif dset == 'synthetic':
dataset = loadSyntheticData()
else:
raise ValueError('Invalid dataset: ' + dset)
return dataset
import os, time, sys
sys.path.append('../')
import numpy as np
from datasets.load import loadDataset
from parse_args_dkf import params
from utils.misc import removeIfExists, createIfAbsent, mapPrint, saveHDF5, displayTime
params['dim_stochastic'] = 1
if params['dataset'] == '':
params['dataset'] = 'synthetic9'
dataset = loadDataset(params['dataset'])
params['savedir'] += '-' + params['dataset']
createIfAbsent(params['savedir'])
#Saving/loading
for k in ['dim_observations', 'dim_actions', 'data_type']:
params[k] = dataset[k]
mapPrint('Options: ', params)
#Setup VAE Model (or reload from existing savefile)
start_time = time.time()
from stinfmodel.dkf import DKF
import stinfmodel.evaluate as DKF_evaluate
import stinfmodel.learning as DKF_learn
displayTime('import DKF', start_time, time.time())
dkf = None
#Remove from params
start_time = time.time()
removeIfExists('./NOSUCHFILE')
import os, time, sys
sys.path.append('../')
import numpy as np
from datasets.load import loadDataset
from optvaedatasets.load import loadDataset as loadDataset_OVAE
from optvaeutils.parse_args import params
from utils.misc import removeIfExists, createIfAbsent, mapPrint, saveHDF5, displayTime, getLowestError
from sklearn.feature_extraction.text import TfidfTransformer
dataset = params['dataset']
params['savedir'] += '-' + dataset + '-' + params['opt_type']
createIfAbsent(params['savedir'])
if 'mnist' in dataset:
dataset = loadDataset(dataset)
if 'binarized' not in dataset:
dataset['train'] = (dataset['train'] > 0.5) * 1.
dataset['valid'] = (dataset['valid'] > 0.5) * 1.
dataset['test'] = (dataset['test'] > 0.5) * 1.
else:
dataset = loadDataset_OVAE(dataset)
#Store dataset parameters into params
for k in ['dim_observations', 'data_type']:
params[k] = dataset[k]
if params['data_type'] == 'bow':
params['max_word_count'] = dataset['train'].max()
mapPrint('Options: ', params)
#Setup VAE Model (or reload from existing savefile)
start_time = time.time()
from optvaemodels.vae import VAE as Model
import optvaemodels.vae_learn as Learn
import optvaemodels.vae_evaluate as Evaluate
lr = lr,
grad_noise = self.params['grad_noise'],
rng = self.srng)#,
self._p('# additional updates: '+str(len(self.updates)))
optimizer_up+=anneal_update +self.updates
fxn_inputs = [X]
self.train = theano.function(fxn_inputs, [elbo, norm_list[0], norm_list[1], norm_list[2], anneal.sum(), lr.sum()],
updates = optimizer_up, name = 'Train')
else:
assert False,'Invalid optimization type: '+self.params['opt_type']
self._p('Done creating functions for training')
if __name__=='__main__':
print 'Initializing VAE'
pfile = './tmp'
from optvaeutils.parse_args_vae import params
params['dim_observations'] =2000
params['data_type'] = 'bow'
params['opt_type'] = 'none'
params['opt_method'] = 'adam'
params['anneal_finopt_rate'] = 100
params['GRADONLY']=True
vae = VAE(params, paramFile=pfile)
from datasets.load import loadDataset
dataset = loadDataset('binarized_mnist')
np.random.seed(1)
idxlist = np.random.permutation(dataset['train'].shape[0])
X = dataset['train'][idxlist[:200]].astype('float32')
os.remove(pfile)
import ipdb;ipdb.set_trace()
Y = labels[st_idx:end_idx][:, None].astype(config.floatX)
batch_nll = self.evaluate(X=X, Y=Y)
nll += batch_nll
self._p(('\tBnum:%d, Batch Bound: %.4f') %
(bnum, batch_nll / float(X.shape[0])))
nll /= float(X.shape[0])
end_time = time.time()
self._p(('(Evaluation) NLL: %.4f [Took %.4f seconds]') %
(nll, end_time - start_time))
return nll
if __name__ == '__main__':
print 'Starting Logistic Regression'
from datasets.load import loadDataset
mnist = loadDataset('mnist')
labels_train = (mnist['train_y'] >= 5.) * 1.
labels_test = (mnist['test_y'] >= 5.) * 1.
params = {}
from utils.parse_args_vae import params
params['dim_observations'] = 784
params['validate_only'] = False
pfile = 'tmp'
LR = LogisticRegression(params, paramFile=pfile)
results = LR.learn(mnist['train'],
labels_train,
epoch_start=0,
epoch_end=100,
savefreq=10,
batch_size=2000,
dataset_eval=mnist['test'],