pl.imshow(oldspec, origin="lower", aspect="auto")
pl.show()
batchdata = np.asarray(test_data, dtype=theano.config.floatX)
delay = 30
# seqlen = [ test_data.shape[0] ] #for multiple files after one another: takes as first input the signal at later time, so delayed training corresponds
hidden_layers_sizes = [10]
numpy_rng = np.random.RandomState(123)
n_dim = [test_data.shape[1]]
dbn_tadbn = TADBN(
numpy_rng=numpy_rng,
n_ins=[n_dim],
hidden_layers_sizes=hidden_layers_sizes,
sparse=0.0,
delay=delay,
learning_rate=0.01,
)
dbn_tadbn.pretrain(
batchdata,
plot_interval=5,
static_epochs=80,
save_interval=10,
ae_epochs=80,
all_epochs=50,
batch_size=5,
seqlen=seqlen,
)
test_data = np.concatenate((test_data,cepstrum))
#print 'time slizes: %i || input dimensions: %i || window size:%i' %(test_data.shape[0],test_data.shape[1],nfft)
batchdata = np.asarray(test_data, dtype=theano.config.floatX)
#seqlen = [ test_data.shape[0] ] #for multiple files after one another: takes as first input the signal at later time, so delayed training corresponds
numpy_rng = np.random.RandomState(123)
n_dim = [test_data.shape[1]]
dbn_tadbn = TADBN(numpy_rng=numpy_rng, n_ins=[n_dim],
hidden_layers_sizes=hidden_layers_sizes,
sparse=sparse, delay=delay, learning_rate=0.01)
dbn_tadbn.pretrain(batchdata, plot_interval=5, static_epochs=80,
save_interval=10, ae_epochs=80, all_epochs=50,
batch_size=5,seqlen=seqlen)
output = open('trained_models/useful/' + savedname + '.pkl', 'wb')
cPickle.dump(dbn_tadbn, output)
output.close()
#===============================================================================
#
# # sanity check
# generated_series = dbn_tadbn.generate(batchdata, n_samples=300)[0,:,:]
# output = open('output/gen_' + savednamed + '.pkl', 'wb')
# cPickle.dump([generated_series,test_data,delay,hidden_layers_sizes,invD,mu,sigma], output)
import numpy as np
import theano
test_data = np.array([np.sin(np.arange(400) * 0.2),
np.sin(np.arange(400) * 0.4)]).T
batchdata = numpy.asarray(test_data, dtype=theano.config.floatX)
delay = 0
numpy_rng = numpy.random.RandomState(123)
n_dim = [test_data.shape[1]]
dbn_tadbn = TADBN(numpy_rng=numpy_rng, n_ins=[n_dim],
hidden_layers_sizes=[100],
sparse=0.0, delay=delay, learning_rate=0.01)
dbn_tadbn.pretrain(batchdata, plot_interval=5, static_epochs=50,
save_interval=10, ae_epochs=0, all_epochs=0,
batch_size=5)
up = dbn_tadbn.propup(batchdata, static=True)
up = np.array(up)
print up
generated_series = dbn_tadbn.generate(batchdata, n_samples=40)
plt.figure()
plt.subplot(211)
import sys
sys.path.append('..')
from models.tadbn import TADBN
import numpy
import numpy as np
import theano
test_data = np.array([np.sin(np.arange(400) * 0.2),
np.sin(np.arange(400) * 0.4)]).T
batchdata = numpy.asarray(test_data, dtype=theano.config.floatX)
delay = 3
numpy_rng = numpy.random.RandomState(123)
n_dim = [test_data.shape[1]]
dbn_tadbn = TADBN(numpy_rng=numpy_rng, n_ins=[n_dim],
hidden_layers_sizes=[10],
sparse=0.0, delay=delay, learning_rate=0.01)
up = dbn_tadbn.propup(batchdata, static=False)
up = np.array(up)
print up.shape, batchdata.shape
