if pre_test_probs == None:
pre_test_probs = prmlps[i].pretrain_test_probs
else:
pre_test_probs = numpy.column_stack((pre_test_probs, prmlps[i].pretrain_test_probs))
print "In the end the test score is %f " % (numpy.mean(test_scores))
return pre_test_probs
if __name__=="__main__":
x = T.matrix('x')
dataset = "/data/lisa/data/pentomino/pentomino64x64_4k_pre.npy"
ds = Dataset()
print "starting pretrain"
ds.setup_pretraining_dataset(data_path=dataset)
train_set_patches, train_set_pre, train_set_labels = ds.Xtrain_patches, ds.Xtrain_presences, ds.Ytrain
test_set_patches, test_set_pre, test_set_labels = ds.Xtest_patches, ds.Xtest_presences, ds.Ytest
prmlps = [PreMLP(x) for each in xrange(64)]
print "starting pretest"
pre_train_probs = train_prmlp(prmlps, train_set_patches, train_set_pre)
pre_test_probs = test_prmlp(prmlps, test_set_patches, test_set_pre)
post_mlp = PosttrainMLP(x, n_in=64*11, n_hidden=200, n_out=10)
post_mlp.posttrain(data=pre_train_probs, labels=train_set_labels, batch_size=80)
post_mlp.posttest(data=pre_test_probs, labels=test_set_labels)
print "starting pretrain"
ds.setup_pretraining_dataset(data_path=dataset)
train_set_patches, train_set_pre, train_set_labels = ds.Xtrain_patches, ds.Xtrain_presences, ds.Ytrain
test_set_patches, test_set_pre, test_set_labels = ds.Xtest_patches, ds.Xtest_presences, ds.Ytest
prmlp = PreMLP(x, n_epochs=2)
post_mlp = PosttrainMLP(x, n_in=64 * 11, n_hidden=400, n_out=10)
print "starting pre-training"
pre_train_probs = train_prmlp(prmlp, train_set_patches, train_set_pre)
print "starting the pre-testing"
pre_test_probs = test_prmlp(prmlp, test_set_patches, test_set_pre)
print "starting post-training"
post_mlp.posttrain(learning_rate=0.001,
data=pre_train_probs,
n_epochs=4,
labels=train_set_labels,
batch_size=60,
save_costs_file=True,
cost_type="negativelikelihood")
print "starting post-testing"
post_mlp.posttest(data=pre_test_probs,
labels=test_set_labels,
save_costs_file=True)
print "starting pretrain"
ds.setup_pretraining_dataset(data_path=dataset)
train_set_patches, train_set_pre, train_set_labels = ds.Xtrain_patches, ds.Xtrain_presences, ds.Ytrain
test_set_patches, test_set_pre, test_set_labels = ds.Xtest_patches, ds.Xtest_presences, ds.Ytest
prmlp = PreMLP(x, n_epochs=2)
post_mlp = PosttrainMLP(x, n_in=64 * 11, n_hidden=400, n_out=10)
print "starting pre-training"
pre_train_probs = train_prmlp(prmlp, train_set_patches, train_set_pre)
print "starting the pre-testing"
pre_test_probs = test_prmlp(prmlp, test_set_patches, test_set_pre)
print "starting post-training"
post_mlp.posttrain(
learning_rate=0.001,
data=pre_train_probs,
n_epochs=4,
labels=train_set_labels,
batch_size=60,
save_costs_file=True,
cost_type="negativelikelihood",
)
print "starting post-testing"
post_mlp.posttest(data=pre_test_probs, labels=test_set_labels, save_costs_file=True)