def return_grad(test_params, input_x, truth_y):
tmp = get_params(model_ft.models_stack[-1])
set_params(model_ft.models_stack[-1], test_params)
result = numpy.concatenate(
[numpy.array(i).flatten() for i in fun_grad(input_x, truth_y)])
set_params(model_ft.models_stack[-1], tmp)
return result
def return_grad(test_params, input_x, truth_y):
tmp = get_params(model_ft.models_stack[-1])
set_params(model_ft.models_stack[-1], test_params)
result = numpy.concatenate([numpy.array(i).flatten() for i in fun_grad(input_x, truth_y)])
set_params(model_ft.models_stack[-1], tmp)
return result
def return_cost(test_params, input_x, truth_y):
tmp = get_params(model_ft.models_stack[-1])
set_params(model_ft.models_stack[-1], test_params)
result = fun_cost(input_x, truth_y)
set_params(model_ft.models_stack[-1], tmp)
return result
fun_grad = theano.function(
[model_ft.varin, model_ft.models_stack[-1].vartruth],
T.grad(model_ft.models_stack[-1].cost() + model_ft.models_stack[-1].weightdecay(weightdecay),
model_ft.models_stack[-1].params)
)
def return_grad(test_params, input_x, truth_y):
tmp = get_params(model_ft.models_stack[-1])
set_params(model_ft.models_stack[-1], test_params)
result = numpy.concatenate([numpy.array(i).flatten() for i in fun_grad(input_x, truth_y)])
set_params(model_ft.models_stack[-1], tmp)
return result
p, g, numlinesearches = minimize(
get_params(model_ft.models_stack[-1]), return_cost, return_grad,
(train_x.get_value(), train_y.get_value()), logreg_epc, verbose=False
)
set_params(model_ft.models_stack[-1], p)
save_params(model_ft, 'ZLIN_4000_1000_4000_1000_4000_1000_4000_10_normhid_nolinb_cae1_dropout.npy')
print "***error rate: train: %f, test: %f" % (
train_set_error_rate(), test_set_error_rate()
)
#############
# FINE-TUNE #
#############
"""
print "\n\n... fine-tuning the whole network"
truth = T.lmatrix('truth')
trainer = GraddescentMinibatch(
varin=model_ft.varin, data=train_x,
truth=model_ft.models_stack[-1].vartruth, truth_data=train_y,
def return_cost(test_params, input_x, truth_y):
tmp = get_params(model_ft.models_stack[-1])
set_params(model_ft.models_stack[-1], test_params)
result = fun_cost(input_x, truth_y)
set_params(model_ft.models_stack[-1], tmp)
return result
def return_grad(test_params, input_x, truth_y):
tmp = get_params(model_ft.models_stack[-1])
set_params(model_ft.models_stack[-1], test_params)
result = numpy.concatenate(
[numpy.array(i).flatten() for i in fun_grad(input_x, truth_y)])
set_params(model_ft.models_stack[-1], tmp)
return result
p, g, numlinesearches = minimize(get_params(model_ft.models_stack[-1]),
return_cost,
return_grad,
(train_x.get_value(), train_y.get_value()),
logreg_epc,
verbose=False)
set_params(model_ft.models_stack[-1], p)
save_params(
model_ft,
'ZLIN_4000_1000_4000_1000_4000_1000_4000_10_normhid_nolinb_cae1_dropout.npy'
)
print "***error rate: train: %f, test: %f" % (train_set_error_rate(),
test_set_error_rate())
#############
# FINE-TUNE #
#############
"""
print "\n\n... fine-tuning the whole network"
truth = T.lmatrix('truth')
trainer = GraddescentMinibatch(
varin=model_ft.varin, data=train_x,
from copy import deepcopy
from master_file import regret_dgrind, regret_exp3
from params import set_params
import math
num_repetitions = 30
dgrind = []
exp3 = []
min_num_rounds = 0
max_num_rounds = 1000
step = 5
rounds = [T for T in range(min_num_rounds, max_num_rounds)] #size: 5000 bytes
T = max_num_rounds
(num_agents, dim, x_real, calA, agent_types, true_labels, delta, noise,
p) = set_params(T)
cp_xreal = deepcopy(x_real)
for delta in [0.05, 0.10, 0.15, 0.3, 0.5]:
print("Current delta = %.5f" % delta)
agents_dgrind = [Agent(t, agent_types, cp_xreal, delta) for t in range(T)]
oracle_dgrind = Oracle(deepcopy(agents_dgrind), calA, T)
resp_lst_dgrind = oracle_dgrind.compute_responses(dim)
print("Independence Number")
# computes approximate independence number in order to tune \eta for GRINDER correctly
a_G = oracle_dgrind.compute_independence_number(dim, resp_lst_dgrind)
print a_G
principal_dgrind = [
Principal(T, calA, num_repetitions, p, a_G)
for _ in range(0, num_repetitions)
from copy import deepcopy
from master_file import regret_grind, regret_exp3
from params import set_params
import math
num_repetitions = 30
dgrind = []
exp3 = []
min_num_rounds = 0
max_num_rounds = 1000
step = 5
rounds = [T for T in range(min_num_rounds, max_num_rounds)]
T = max_num_rounds
(num_agents, dim, x_real, calA_exp3, calA_grind, agent_types, true_labels,
delta, noise, prob) = set_params(T, 0.2)
cp_xreal = deepcopy(x_real)
for delta in [0.05, 0.1, 0.15, 0.3, 0.5]:
print("Current delta = %.5f" % delta)
agents_grind = [Agent(t, agent_types, cp_xreal, delta) for t in range(T)]
oracle_grind = Oracle(deepcopy(agents_grind), T)
principal_grind = [
Principal_Grind(T, calA_grind, num_repetitions)
for _ in range(0, num_repetitions)
]
principal_exp3 = [
Principal_Exp3(T, calA_exp3, num_repetitions, 0)
for _ in range(0, num_repetitions)
]