def get_profiler(profiling, lang='c'):
""" Figures out which profiling linker to use """
if profiling:
if lang == 'c':
profmode = theano.ProfileMode(optimizer='fast_run',
linker=theano.gof.OpWiseCLinker())
elif lang == 'py':
profmode = theano.ProfileMode(optimizer='fast_run',
linker=theano.gof.PerformLinker())
else:
profmode = None
return profmode
def test_module_advanced_example():
profmode = theano.ProfileMode(optimizer='fast_run',
linker=theano.gof.OpWiseCLinker())
profmode = PrintEverythingMode(theano.gof.OpWiseCLinker(), 'fast_run')
data_x = N.random.randn(4, 10)
data_y = [[int(x)] for x in (N.random.randn(4) > 0)]
model = SpecifiedRegressionLayer(regularize=False).make(input_size=10,
target_size=1,
stepsize=0.1,
mode=profmode)
for i in xrange(1000):
xe, gw, gb, ga = model.update(data_x, data_y)
if i % 100 == 0:
print(i, xe)
pass
#for inputs, targets in my_training_set():
#print "cost:", model.update(inputs, targets)
print("final weights:", model.w)
print("final biases:", model.b)
profmode.print_summary()
def __init__(self,
network,
profile=False,
lr=0.3,
L1=0,
momentum=0.4,
L2=0.00005,
compile=True):
self.profile = profile
self.network = network
self.lr = np.float32(lr)
self.momentum = np.float32(momentum)
self.params = network.params
self.cost = network.cost + np.float32(L2) * T.sum([
(p**2).sum() for p in self.params
]) + np.float32(L1) * T.sum([abs(p).sum() for p in self.params])
self.grads = T.grad(self.cost, self.params)
# Expressions evaluated for training
self.cost_exprs = [self.cost, network.cost]
self.cost_names = ['L2 cost', "Network cost"]
for name, monitor in network.monitors:
self.cost_names.append(name)
self.cost_exprs.append(monitor)
self.shapes = [p.get_value(borrow=True).shape for p in self.params]
self.counts = [np.prod(s) for s in self.shapes]
self.starts = np.cumsum([0] + self.counts)[:-1]
self.dtype = self.params[0].get_value().dtype
self.best_cost = 1e100
self.best_iter = 0
self.best_params = [p.get_value().copy() for p in self.params]
mode = "FAST_RUN"
if self.profile:
mode = theano.ProfileMode(optimizer='fast_run',
linker=theano.gof.OpWiseCLinker())
if compile:
self.f_eval = theano.function(network.inputs, network.output)
self.f_learn = theano.function(network.inputs + [network.target],
self.cost_exprs,
updates=list(
self.learning_updates()),
mode=mode)
self.f_cost = theano.function(network.inputs + [network.target],
self.cost_exprs)
import numpy as np
import gzip, cPickle
import matplotlib.pyplot as plt
plt.ion()
from utils import generate_data, get_context
# DEBUGGING
from theano import ProfileMode
mode = theano.ProfileMode(optimizer='fast_run',
linker=theano.gof.OpWiseCLinker())
# mode = theano.compile.DebugMode(check_py_code=False, require_matching_strides=False)
# mode = None
# load data
print ">> Loading dataset..."
f = gzip.open('datasets/mnist.pkl.gz', 'rb')
train_set, valid_set, test_set = cPickle.load(f)
f.close()
train_set_x, train_set_y = train_set
valid_set_x, valid_set_y = valid_set
test_set_x, test_set_y = test_set
# TODO DEBUG
def get_profile_mode():
profmode = theano.ProfileMode(optimizer='fast_run', linker=theano.gof.OpWiseCLinker())
return profmode
import numpy as np
import theano
import theano.tensor as T
from nn.math import softmax, sigmoid, make_onehot
from theano import pp
import pandas as pd
from data_utils import utils as du
from theano import ProfileMode
import theano.tensor.nnet as nn
from misc import *
from nn.base import NNBase
profmode = theano.ProfileMode(optimizer='fast_compile')
dtype = theano.config.floatX
N_ASPECTS = 5
N_SENTIMENTS = 3
#n_hidden = 50
# squashing of the gates should result in values between 0 and 1
# therefore we use the logistic function
sigma = lambda x: 1 / (1 + T.exp(-x))
# for the other activation function we use the tanh
act = T.tanh
def one_lstm_step(seq, seq2, h_tm1, h_tm2, c_tm1, c_tm2, coeff, coeff_rev,
W_hy, b_y, W_xi, W_xi2, U_hi, U_hi2, b_i, b_i2, W_xf, W_xf2,
U_hf, U_hf2, b_f, b_f2, W_xc, W_xc2, U_hc, U_hc2, b_c, b_c2,