if __name__ == '__main__':
params = config.load_parameters(sys.argv[1])
dataset = load_data(params.dataset,
resize_to = params.resize_data_to,
shared = False,
pickled = params.pickled)
x = T.matrix('x')
y = T.ivector('y')
index = T.lscalar('index')
method = params.method
method.prepare(params,dataset)
train_set = method.resampler.get_train()
valid_set = method.resampler.get_valid()
test_set = method.resampler.get_test()
test_set_x, test_set_y = test_set
shared_dataset = [train_set,valid_set,test_set]
continuations = dill.load(open(sys.argv[2]))
members = []
i = 0
for c in continuations:
print "training member {0}".format(i)
m = test_mlp(shared_dataset, params, continuation = c, x=x, y=y,
index=index)
members.append(m.get_weights())
m.clear()
del m
gc.collect()
i += 1
dill.dump(members,open(sys.argv[3],"wb"))
print "saving results to {0}@{1}".format(params.results_db, host)
conn = MongoClient(host=host)
db = conn[params.results_db]
if 'results_table' in params.__dict__:
table_name = params.results_table
else:
table_name = 'results'
table = db[table_name]
results = {
"params": params.__dict__,
"test_losses": test_losses,
"test_score": test_score,
}
table.insert(json.loads(json.dumps(results, default=common.serialize)))
train, valid, test = dataset
train_x, train_y = train
valid_x, valid_y = valid
test_x, test_y = test
shared_train_x = sharedX(train_x)
shared_valid_x = sharedX(valid_x)
shared_test_x = sharedX(test_x)
distilled_train_y = ensemble.classify(shared_train_x).eval()
distilled_set = (
(shared_train_x, sharedX(distilled_train_y, dtype=numpy.int32)),
(shared_valid_x, sharedX(valid_y, dtype=numpy.int32)),
(shared_test_x, sharedX(test_y, dtype=numpy.int32)),
)
#y = T.vector()
mlp = test_mlp(distilled_set, params, y=y)
arg_param_pairings = [
(args.seed, 'random_seed'),
(args.results_db, 'results_db'),
(args.results_host, 'results_host'),
(args.results_table, 'results_table'),
(args.epochs, 'n_epochs'),
]
from toupee import config
params = config.load_parameters(args.params_file)
def arg_params(arg_value, param):
if arg_value is not None:
params.__dict__[param] = arg_value
for arg, param in arg_param_pairings:
arg_params(arg, param)
from toupee import data
from toupee.mlp import MLP, test_mlp
dataset = data.load_data(params.dataset,
resize_to=params.resize_data_to,
shared=False,
pickled=params.pickled,
center_and_normalise=params.center_and_normalise,
join_train_and_valid=params.join_train_and_valid)
pretraining_set = data.make_pretraining_set(dataset, params.pretraining)
mlp = test_mlp(dataset, params, pretraining_set=pretraining_set)
if args.save_file is not None:
dill.dump(mlp, open(args.save_file, "wb"))
arg_param_pairings = [
(args.seed, 'random_seed'),
(args.results_db, 'results_db'),
(args.results_host, 'results_host'),
(args.results_table, 'results_table'),
(args.epochs, 'n_epochs'),
]
from toupee import config
params = config.load_parameters(args.params_file)
def arg_params(arg_value,param):
if arg_value is not None:
params.__dict__[param] = arg_value
for arg, param in arg_param_pairings:
arg_params(arg,param)
from toupee import data
from toupee.mlp import MLP, test_mlp
dataset = data.load_data(params.dataset,
resize_to = params.resize_data_to,
shared = False,
pickled = params.pickled,
center_and_normalise = params.center_and_normalise,
join_train_and_valid = params.join_train_and_valid)
pretraining_set = data.make_pretraining_set(dataset,params.pretraining)
mlp = test_mlp(dataset, params, pretraining_set = pretraining_set)
if args.save_file is not None:
dill.dump(mlp,open(args.save_file,"wb"))
params.pretraining = None
dataset = load_data(params.dataset,
resize_to = params.resize_data_to,
shared = False,
pickled = params.pickled)
x = T.matrix('x')
y = T.ivector('y')
index = T.lscalar('index')
method = params.method
method.prepare(params,dataset)
train_set = method.resampler.get_train()
valid_set = method.resampler.get_valid()
test_set = method.resampler.get_test()
shared_dataset = [train_set,valid_set,test_set]
continuations = dill.load(open(sys.argv[2]))
members = [test_mlp(shared_dataset, params, continuation = c, x=x, y=y,
index=index) for c in continuations]
ensemble = params.method.create_aggregator(params,members,x,y,train_set,valid_set)
test_set_x, test_set_y = method.resampler.get_test()
test_model = theano.function(inputs=[index],
outputs=ensemble.errors,
givens={
x: test_set_x[index * params.batch_size:(index + 1) *
params.batch_size],
y: test_set_y[index * params.batch_size:(index + 1) *
params.batch_size]})
n_test_batches = test_set_x.shape[0].eval() / params.batch_size
test_losses = [test_model(i) for i in xrange(n_test_batches)]
test_score = numpy.mean(test_losses)
print 'Final error: {0} %'.format(test_score * 100.)
if 'results_table' in params.__dict__:
table_name = params.results_table
else:
table_name = 'results'
table = db[table_name]
results = {
"params": params.__dict__,
"test_losses" : test_losses,
"test_score" : test_score,
}
table.insert(json.loads(json.dumps(results,default=common.serialize)))
train,valid,test = dataset
train_x,train_y = train
valid_x,valid_y = valid
test_x,test_y = test
shared_train_x = sharedX(train_x)
shared_valid_x = sharedX(valid_x)
shared_test_x = sharedX(test_x)
distilled_train_y = ensemble.classify(shared_train_x).eval()
distilled_set = (
(shared_train_x,
sharedX(distilled_train_y, dtype = numpy.int32)),
(shared_valid_x,
sharedX(valid_y, dtype = numpy.int32)),
(shared_test_x,
sharedX(test_y, dtype = numpy.int32)),
)
#y = T.vector()
mlp = test_mlp(distilled_set, params, y = y)