def test(expid):
# to test:
# OMP_NUM_THREADS=1 THEANO_FLAGS=device=cpu taskset -c 0 python $(expip)/code.py $(expid)
import os
os.chdir(expid)
print "loading data"
data = GenericClassificationDataset("cifar10", "../cifar_10_shuffled.pkl")
print "building model"
model, learn, test = build_model(False)
print "importing weights"
shared.importFromFile("weights.pkl")
print "testing"
import time
t0 = time.time()
test_error, test_cost = data.doTest(test, 50)
t1 = time.time()
print "Error, cost, time(s)"
print test_error, test_cost, t1 - t0
specialized_test_time = t1 - t0
normal_test_time = t1 - t0
f = file("test_results.txt", 'w')
f.write("specialized:%f\ntheano:%f\nerror:%f\n" %
(specialized_test_time, normal_test_time, test_error))
f.close()
def test(expid):
# to test:
# OMP_NUM_THREADS=1 THEANO_FLAGS=device=cpu taskset -c 0 python $(expip)/code.py $(expid)
import os
os.chdir(expid)
print "loading data"
data = GenericClassificationDataset("cifar10", "../cifar_10_shuffled.pkl")
print "building model"
model,learn,test = build_model(False)
print "importing weights"
shared.importFromFile("weights.pkl")
print "testing"
import time
t0 = time.time()
test_error, test_cost = data.doTest(test, 50)
t1 = time.time()
print "Error, cost, time(s)"
print test_error, test_cost, t1-t0
specialized_test_time = t1-t0
normal_test_time = t1-t0
f= file("test_results.txt",'w')
f.write("specialized:%f\ntheano:%f\nerror:%f\n"%(specialized_test_time, normal_test_time, test_error))
f.close()
def main():
data = GenericClassificationDataset("cifar10", "cifar_10_shuffled.pkl")
N = data.train[0].shape[0] * 1.
model, learn, test = build_model()
some_probs = []
epoch = 0
experiment = {"results":None,
}
lr = 0.001 # * 100 / (i+100)
costs = []
errors = []
valid_costs = []
valid_errors = []
for i in range(1000):
epoch = i
cost = 0
error = 0
for x,y in data.trainMinibatches(128):
c,e = learn(x,y,lr)
cost += c
error += e
t0 = time.time()
valid_error, valid_cost = data.validate(test, 50)
valid_time = time.time() - t0
print
print i, cost/N, error/N
print valid_error, valid_cost, valid_time
errors.append(error/N)
costs.append(cost/N)
valid_errors.append(valid_error)
valid_costs.append(valid_cost)
tools.export_feature_image(model.layers[0].h.W, "W_img.png", (32,32,3))
tools.export_multi_plot1d([errors, valid_errors], "errors.png", "error")
tools.export_multi_plot1d([costs, valid_costs], "costs.png", "cost")
experiment["results"] = [valid_costs, valid_errors, costs, errors]
experiment["valid_time"] = valid_time
pickle.dump(experiment, file("experiment.pkl",'w'),-1)
shared.exportToFile("weights.pkl")
def main():
data = GenericClassificationDataset("cifar10", "cifar_10_shuffled.pkl")
N = data.train[0].shape[0] * 1.
model, learn, test = build_model()
some_probs = []
epoch = 0
experiment = {
"results": None,
}
lr = 0.001 # * 100 / (i+100)
costs = []
errors = []
valid_costs = []
valid_errors = []
for i in range(1000):
epoch = i
cost = 0
error = 0
for x, y in data.trainMinibatches(128):
c, e = learn(x, y, lr)
cost += c
error += e
t0 = time.time()
valid_error, valid_cost = data.validate(test, 50)
valid_time = time.time() - t0
print
print i, cost / N, error / N
print valid_error, valid_cost, valid_time
errors.append(error / N)
costs.append(cost / N)
valid_errors.append(valid_error)
valid_costs.append(valid_cost)
tools.export_feature_image(model.layers[0].h.W, "W_img.png",
(32, 32, 3))
tools.export_multi_plot1d([errors, valid_errors], "errors.png",
"error")
tools.export_multi_plot1d([costs, valid_costs], "costs.png", "cost")
experiment["results"] = [valid_costs, valid_errors, costs, errors]
experiment["valid_time"] = valid_time
pickle.dump(experiment, file("experiment.pkl", 'w'), -1)
shared.exportToFile("weights.pkl")
def main():
data = GenericClassificationDataset("cifar10", "cifar_10_shuffled.pkl")
N = data.train[0].shape[0] * 1.
model, learn, test = build_model()
do_video = True
some_probs = []
fps = 30
try:
video = Popen(['avconv', '-y', '-f', 'image2pipe', '-vcodec', 'mjpeg',#'rawvideo', "-pix_fmt", "rgba",
'-r', str(fps),'-s','800x600', '-i', '-',
'-qscale', '9', '-r', str(fps), 'video.webm'], stdin=PIPE)
except Exception,e:
print "Cannot do video:",e
do_video = False
