def predict(self, X, var=False, max_rows=1000):
"""Predict the target values.
Parameters
----------
X : array_like
Array of shape (n_samples, n_features).
var : boolean
Flag indicating whether the variance of the predictions should be
returned as well. In this case, the complexity rises from O(n) to
O(n^2).
max_rows : integer
Maximum number of predictions to do in one step; a lower number
might help performance.
Returns
-------
mean : array_like
Array of the form (n_samples, 1) containing the mean of the
predictions.
variance : array_like
Only if ``var == True``. Array of shape (n_samples, 1) containing
the variance of the predictions.
"""
if self.f_predict is None or self.f_predict_var is None:
self.f_predict, self.f_predict_var = self._make_predict_functions(self.stored_X, self.stored_Z)
n_steps, rest = divmod(X.shape[0], max_rows)
if rest != 0:
n_steps += 1
steps = [(i * max_rows, (i + 1) * max_rows) for i in range(n_steps)]
X = (X - self.mean_x) / self.std_x
if var:
Y, = theano_floatx(np.empty((X.shape[0], 1)))
Y_var, = theano_floatx(np.empty((X.shape[0], 1)))
for start, stop in steps:
this_x = X[start:stop]
m, s = self.f_predict_var(this_x)
Y[start:stop] = m
Y_var[start:stop] = s
Y = (Y * self.std_z) + self.mean_z
Y_var = Y_var * self.std_z
return Y, Y_var
else:
Y, = theano_floatx(np.empty((X.shape[0], 1)))
for start, stop in steps:
this_x = X[start:stop]
Y[start:stop] = self.f_predict(this_x)
Y = (Y * self.std_z) + self.mean_z
return Y
def _make_start_exprs(self):
exprs = {
'inpt': T.tensor3('inpt')
}
exprs['inpt'].tag.test_value, = theano_floatx(np.ones((3, 2, self.n_inpt)))
if self.imp_weight:
exprs['imp_weight'] = T.tensor3('imp_weight')
exprs['imp_weight'].tag.test_value, = theano_floatx(np.ones((3, 2, 1)))
return exprs
def _make_start_exprs(self):
exprs = {'inpt': T.tensor3('inpt')}
exprs['inpt'].tag.test_value, = theano_floatx(
np.ones((3, 2, self.n_inpt)))
if self.imp_weight:
exprs['imp_weight'] = T.tensor3('imp_weight')
exprs['imp_weight'].tag.test_value, = theano_floatx(
np.ones((3, 2, 1)))
return exprs
def _make_start_exprs(self):
inpt = T.tensor3('inpt')
inpt.tag.test_value, = theano_floatx(np.ones((4, 3, self.n_inpt)))
if self.use_imp_weight:
imp_weight = T.tensor3('imp_weight')
imp_weight.tag.test_value, = theano_floatx(np.ones((4, 3, 1)))
else:
imp_weight = None
return inpt, imp_weight
def _make_start_exprs(self):
inpt = T.matrix('inpt')
inpt.tag.test_value, = theano_floatx(np.ones((3, self.n_inpt)))
if self.use_imp_weight:
imp_weight = T.matrix('imp_weight')
imp_weight.tag.test_value, = theano_floatx(np.ones((3, 1)))
else:
imp_weight = None
return inpt, imp_weight
def _make_start_exprs(self):
inpt = T.tensor3("inpt")
inpt.tag.test_value, = theano_floatx(np.ones((4, 3, self.n_inpt)))
if self.use_imp_weight:
imp_weight = T.tensor3("imp_weight")
imp_weight.tag.test_value, = theano_floatx(np.ones((4, 3, 1)))
else:
imp_weight = None
return inpt, imp_weight
def test_gp_fit_linear():
X = np.arange(-2, 2, .01)[:, np.newaxis].astype(theano.config.floatX)
X, = theano_floatx(X)
idxs = range(X.shape[0])
idxs = random.sample(idxs, 200)
X = X[idxs]
Z = np.sin(X)
X, Z = theano_floatx(X, Z)
gp = GaussianProcess(1, max_iter=10, kernel='linear')
gp.fit(X, Z)
def test_gp_fit_linear():
X = np.arange(-2, 2, .1)[:, np.newaxis].astype(theano.config.floatX)
X, = theano_floatx(X)
idxs = range(X.shape[0])
idxs = random.sample(idxs, 20)
X = X[idxs]
Z = np.sin(X)
X, Z = theano_floatx(X, Z)
gp = GaussianProcess(1, max_iter=10, kernel='linear')
gp.fit(X, Z)
def _make_start_exprs(self):
inpt = T.tensor4('inpt')
inpt.tag.test_value, = theano_floatx(np.ones(
(3, self.n_channel, self.image_height, self.image_width)))
if self.use_imp_weight:
imp_weight = T.tensor4('imp_weight')
imp_weight.tag.test_value, = theano_floatx(np.ones(
(3, self.n_channel, self.image_height, self.image_width)))
else:
imp_weight = None
return inpt, imp_weight
def _make_start_exprs(self):
inpt = T.tensor4('inpt')
inpt.tag.test_value, = theano_floatx(
np.ones((3, self.n_channel, self.image_height, self.image_width)))
if self.use_imp_weight:
imp_weight = T.tensor4('imp_weight')
imp_weight.tag.test_value, = theano_floatx(
np.ones(
(3, self.n_channel, self.image_height, self.image_width)))
else:
imp_weight = None
return inpt, imp_weight
def test_dmlp_predict():
X = np.random.standard_normal((10, 2))
X, = theano_floatx(X)
mlp = DropoutMlp(2, [10], 1, ['tanh'], 'identity', 'squared', max_iter=10,
p_dropout_inpt=.2, p_dropout_hiddens=[0.5])
mlp.predict(X)
def test_sparse_coding_fit():
raise SkipTest()
X = np.random.standard_normal((10, 2))
X, = theano_floatx(X)
sf = SparseCoding(2, 7, max_iter=10)
sf.fit(X)
def test_rim_iter_fit():
X = np.random.standard_normal((10, 2))
X, = theano_floatx(X)
rim = Rim(2, 10, 0.1, max_iter=10)
for i, info in enumerate(rim.iter_fit(X)):
if i >= 10:
break
def test_storn():
theano.config.compute_test_value = 'raise'
X = np.random.random((3, 5, 2))
X, = theano_floatx(X)
class Assmptn(sgvb.DiagGaussLatentAssumption, sgvb.DiagGaussVisibleAssumption):
pass
m = sgvb.StochasticRnn(
2, [5], 4, [5],
['tanh'] * 1, ['rectifier'] * 1,
assumptions=Assmptn(),
optimizer='rprop', batch_size=None,
max_iter=3)
print 'init pars and expressions'
m._init_pars()
m._init_exprs()
print 'fitting'
m.fit(X)
print 'scoring'
m.score(X)
print 'transforming'
m.transform(X)
print 'sampling map'
m.sample(5, visible_map=True)
print 'sampling'
m.sample(5, visible_map=False)
def test_cnn_predict():
X = np.random.standard_normal((10, 2 * 100 * 50))
X, = theano_floatx(X)
m = Cnn(
100 * 50,
[10, 15],
[20, 12],
1,
['sigmoid', 'sigmoid'],
['rectifier', 'rectifier'],
'sigmoid',
'cat_ce',
100,
50,
2,
optimizer=('rmsprop', {
'step_rate': 1e-4,
'decay': 0.9
}),
batch_size=2,
max_iter=10,
pool_shapes=[(2, 2), (2, 2)],
filter_shapes=[(4, 4), (3, 3)],
)
m.predict(X)
def test_cnn_iter_fit():
X = np.random.standard_normal((10, 2 * 100 * 50))
Z = np.random.random((10, 1)) > 0.5
X, Z = theano_floatx(X, Z)
m = Cnn(
100 * 50,
[10, 15],
[20, 12],
1,
['sigmoid', 'sigmoid'],
['rectifier', 'rectifier'],
'sigmoid',
'cat_ce',
100,
50,
2,
optimizer=('rmsprop', {
'step_rate': 1e-4,
'decay': 0.9
}),
batch_size=2,
max_iter=10,
pool_shapes=[(2, 2), (2, 2)],
filter_shapes=[(4, 4), (3, 3)],
)
for i, info in enumerate(m.iter_fit(X, Z)):
if i >= 10:
break
def test_storn():
X = np.random.random((3, 3, 2))
X, = theano_floatx(X)
kwargs = {
'n_inpt': X.shape[2],
'n_hiddens_recog': [5],
'n_latent': 11,
'n_hiddens_gen': [7],
'recog_transfers': ['tanh'],
'gen_transfers': ['rectifier'],
'p_dropout_inpt': .1,
'p_dropout_hiddens': [.1],
'p_dropout_shortcut': [.1],
'p_dropout_hidden_to_out': .1,
'use_imp_weight': False,
'optimizer': 'adam',
'batch_size': None,
'verbose': False,
'max_iter': 3,
}
m = MyStorn(**kwargs)
print 'fitting'
m.fit(X)
print 'scoring'
m.score(X)
print 'initializing'
m.initialize()
def test_lenet():
image_height, image_width = 16, 16
X = np.random.standard_normal((11, 1, image_height, image_width))
Z = np.random.random((11, 1)) > 0.5
X, Z = theano_floatx(X, Z)
n_hiddens_conv = [5, 2]
filter_shapes = [(2, 2), (2, 2)]
pool_shapes = [(2, 2), (2, 2)]
n_hiddens_full = [20]
transfers_conv = ['tanh', 'tanh']
transfers_full = ['rectifier']
n_channel = 1
n_output = 1
out_transfer = 'identity'
loss = 'squared'
m = Lenet(image_height, image_width, n_channel, n_hiddens_conv,
filter_shapes, pool_shapes, n_hiddens_full, n_output,
transfers_conv, transfers_full, out_transfer, loss)
f_predict = m.function(['inpt'], 'output', mode='FAST_COMPILE')
f_predict(X)
m.fit(X, Z)
def test_sparse_coding_iter_fit():
X = np.random.standard_normal((10, 2))
X, = theano_floatx(X)
sf = SparseCoding(2, 7, max_iter=10)
for i, info in enumerate(sf.iter_fit(X)):
if i >= 10:
break
def test_tsne():
X = np.random.random((100, 3)).astype(theano.config.floatX)
X, = theano_floatx(X)
tsne = Tsne(n_inpt=3, n_lowdim=2, perplexity=40, early_exaggeration=50,
max_iter=10)
E = tsne.fit_transform(X)
def test_rica_reconstruct():
X = np.random.standard_normal((10, 2))
X, = theano_floatx(X)
rica = Rica(2, 10, code_transfer='softabs',
hidden_transfer='identity', loss='squared',
c_ica=0.5, max_iter=10)
rica.reconstruct(X)
def test_minibatch_score_trainer():
X = np.random.random((100, 10))
Z = np.random.random((100, 2))
X, Z = theano_floatx(X, Z)
data = {'train': (X, Z), 'val': (X, Z), 'test': (X, Z)}
cut_size = 10
class MyModel(mlp.Mlp):
def score(self, X, Z):
assert X.shape[0] <= cut_size
return super(MyModel, self).score(X, Z)
m = MyModel(10, [100], 2, ['tanh'], 'identity', 'squared', max_iter=10)
score = MinibatchScore(cut_size, [0, 0])
trainer = Trainer(m,
data,
score=score,
pause=lambda info: True,
stop=lambda info: False)
trainer.val_key = 'val'
for _ in trainer.iter_fit(X, Z):
break
def test_training_continuation():
# Make model and data for the test.
X = np.random.random((10, 2))
X, = theano_floatx(X)
optimizer = 'gd'
m = autoencoder.AutoEncoder(2, [2], ['tanh'],
'identity',
'squared',
tied_weights=True,
max_iter=10,
optimizer=optimizer)
# Train the mdoel with a trainer for 2 epochs.
stopper = climin.stops.OnSignal()
print stopper.sig
stops = climin.stops.Any([stopper, climin.stops.AfterNIterations(5)])
t = Trainer(m, stop=stops, pause=climin.stops.always)
t.val_key = 'val'
t.eval_data = {'val': (X, )}
killed = False
for info in t.iter_fit(X):
os.kill(os.getpid(), stopper.sig)
assert info['n_iter'] == 1
def test_fd_predict():
X = np.random.standard_normal((10, 2))
X, = theano_floatx(X)
loss = lambda target, prediction: squared(target, prediction[:, :target.shape[1]])
mlp = FastDropoutNetwork(
2, [10], 1, ['rectifier'], 'identity', loss, max_iter=10)
mlp.predict(X)
def test_training_continuation():
if sys.platform == 'win32':
raise SkipTest()
# Make model and data for the test.
X = np.random.random((100, 10))
Z = np.random.random((100, 2))
X, Z = theano_floatx(X, Z)
data = {
'train': (X, Z),
'val': (X, Z),
'test': (X, Z)
}
optimizer = 'rmsprop', {'step_rate': 0.0001}
m = mlp.Mlp(10, [2], 2, ['tanh'], 'identity', 'squared', max_iter=10,
optimizer=optimizer)
# Train the mdoel with a trainer for 2 epochs.
stopper = climin.stops.OnSignal()
stops = climin.stops.Any([stopper, climin.stops.AfterNIterations(5)])
t = Trainer(
m,
data,
stop=stops,
pause=climin.stops.always)
t.val_key = 'val'
for info in t.iter_fit(X, Z):
os.kill(os.getpid(), stopper.sig)
assert info['n_iter'] == 1
def test_training_continuation():
# Make model and data for the test.
X = np.random.random((10, 2))
X, = theano_floatx(X)
optimizer = 'gd'
m = autoencoder.AutoEncoder(2, [2], ['tanh'], 'identity', 'squared',
tied_weights=True, max_iter=10,
optimizer=optimizer)
# Train the mdoel with a trainer for 2 epochs.
stopper = climin.stops.OnSignal()
print stopper.sig
stops = climin.stops.Any([stopper, climin.stops.AfterNIterations(5)])
t = Trainer(
m,
stop=stops,
pause=climin.stops.always)
t.val_key = 'val'
t.eval_data = {'val': (X,)}
killed = False
for info in t.iter_fit(X):
os.kill(os.getpid(), stopper.sig)
assert info['n_iter'] == 1
def test_minibatch_score_trainer():
X = np.random.random((100, 10))
X, = theano_floatx(X)
cut_size = 10
class MyAutoEncoder(autoencoder.AutoEncoder):
def score(self, X):
assert X.shape[0] <= cut_size
return super(MyAutoEncoder, self).score(X)
m = MyAutoEncoder(10, [100], ['tanh'],
'identity',
'squared',
tied_weights=True,
max_iter=10)
score = MinibatchScore(cut_size, [0])
trainer = Trainer(m,
score=score,
pause=lambda info: True,
stop=lambda info: False)
trainer.eval_data = {'val': (X, )}
trainer.val_key = 'val'
for _ in trainer.iter_fit(X):
break
def test_minibatch_score_trainer():
X = np.random.random((100, 10))
Z = np.random.random((100, 2))
X, Z = theano_floatx(X, Z)
data = {
'train': (X, Z),
'val': (X, Z),
'test': (X, Z)
}
cut_size = 10
class MyModel(mlp.Mlp):
def score(self, X, Z):
assert X.shape[0] <= cut_size
return super(MyModel, self).score(X, Z)
m = MyModel(10, [100], 2, ['tanh'], 'identity', 'squared',
max_iter=10)
score = MinibatchScore(cut_size, [0, 0])
trainer = Trainer(
m, data, score=score, pause=lambda info: True, stop=lambda info: False)
trainer.val_key = 'val'
for _ in trainer.iter_fit(X, Z):
break
def test_storn():
X = np.random.random((3, 3, 2))
X, = theano_floatx(X)
kwargs = {
'n_inpt': X.shape[2],
'n_hiddens_recog': [5],
'n_latent': 11,
'n_hiddens_gen': [7],
'recog_transfers': ['tanh'],
'gen_transfers': ['rectifier'],
'p_dropout_inpt': .1,
'p_dropout_hiddens': [.1],
'p_dropout_shortcut': [.1],
'p_dropout_hidden_to_out': .1,
'use_imp_weight': False,
'optimizer': 'adam',
'batch_size': None,
'verbose': False,
'max_iter': 3,
}
m = MyStorn(**kwargs)
print 'fitting'
m.fit(X)
print 'scoring'
m.score(X)
print 'initializing'
m.initialize()
def test_training_continuation():
if sys.platform == 'win32':
raise SkipTest()
# Make model and data for the test.
X = np.random.random((100, 10))
Z = np.random.random((100, 2))
X, Z = theano_floatx(X, Z)
data = {'train': (X, Z), 'val': (X, Z), 'test': (X, Z)}
optimizer = 'rmsprop', {'step_rate': 0.0001}
m = mlp.Mlp(10, [2],
2, ['tanh'],
'identity',
'squared',
max_iter=10,
optimizer=optimizer)
# Train the mdoel with a trainer for 2 epochs.
stopper = climin.stops.OnSignal()
stops = climin.stops.Any([stopper, climin.stops.AfterNIterations(5)])
t = Trainer(m, data, stop=stops, pause=climin.stops.always)
t.val_key = 'val'
for info in t.iter_fit(X, Z):
os.kill(os.getpid(), stopper.sig)
assert info['n_iter'] == 1
def test_storn():
theano.config.compute_test_value = 'raise'
X = np.random.random((3, 5, 2))
X, = theano_floatx(X)
class Assmptn(sgvb.DiagGaussLatentAssumption,
sgvb.DiagGaussVisibleAssumption):
pass
m = sgvb.StochasticRnn(2, [5],
4, [5], ['tanh'] * 1, ['rectifier'] * 1,
assumptions=Assmptn(),
optimizer='rprop',
batch_size=None,
max_iter=3)
print 'init pars and expressions'
m._init_pars()
m._init_exprs()
print 'fitting'
m.fit(X)
print 'scoring'
m.score(X)
print 'transforming'
m.transform(X)
print 'sampling map'
m.sample(5, visible_map=True)
print 'sampling'
m.sample(5, visible_map=False)
def test_lde():
X = np.eye(2)
X, = theano_floatx(X)
lde = LinearDenoiser(0.5)
lde.fit(X)
assert np.allclose(lde.weights, [[0.499995, -0.499985], [-0.499985, 0.499995]])
assert np.allclose(lde.bias, [0.499995, 0.499995])
def test_uslstm_iter_fit():
raise SkipTest()
X = np.random.standard_normal((10, 5, 2)).astype(theano.config.floatX)
X, = theano_floatx(X)
rnn = UnsupervisedLstm(2, [10], 3, loss=lambda x: T.log(x), max_iter=10)
for i, info in enumerate(rnn.iter_fit(X)):
if i >= 10:
break
def test_slstm_iter_fit():
X = np.random.standard_normal((10, 5, 2)).astype(theano.config.floatX)
Z = np.random.standard_normal((10, 5, 3)).astype(theano.config.floatX)
X, Z = theano_floatx(X, Z)
rnn = SupervisedLstmRnn(2, [10], 3, hidden_transfers=['sigmoid'], max_iter=10)
for i, info in enumerate(rnn.iter_fit(X, Z)):
if i >= 10:
break
def test_slstm_iter_fit():
X = np.random.standard_normal((10, 5, 2)).astype(theano.config.floatX)
Z = np.random.standard_normal((10, 5, 3)).astype(theano.config.floatX)
X, Z = theano_floatx(X, Z)
rnn = SupervisedLstmRnn(2, [10], 3, hidden_transfers=['sigmoid'], max_iter=10)
for i, info in enumerate(rnn.iter_fit(X, Z)):
if i >= 10:
break
def test_sparse_filtering_iter_fit():
X = np.random.standard_normal((10, 2))
X, = theano_floatx(X)
sf = SparseFiltering(2, 10, max_iter=10)
for i, info in enumerate(sf.iter_fit(X)):
if i >= 10:
break
def test_slstm_predict():
X = np.random.standard_normal((10, 5, 2)).astype(theano.config.floatX)
X, = theano_floatx(X)
rnn = SupervisedLstmRnn(2, [10],
3,
hidden_transfers=['sigmoid'],
max_iter=10)
rnn.predict(X)
def test_autoencoder():
X = np.random.random((10, 10))
X, = theano_floatx(X)
m = autoencoder.AutoEncoder(10, [100], ["tanh"], "identity", "squared", tied_weights=True, max_iter=10)
m.fit(X)
m.score(X)
m.transform(X)
def test_uslstm_iter_fit():
raise SkipTest()
X = np.random.standard_normal((10, 5, 2)).astype(theano.config.floatX)
X, = theano_floatx(X)
rnn = UnsupervisedLstm(2, [10], 3, loss=lambda x: T.log(x), max_iter=10)
for i, info in enumerate(rnn.iter_fit(X)):
if i >= 10:
break
def test_sparse_coding_iter_fit():
raise SkipTest()
X = np.random.standard_normal((10, 2))
X, = theano_floatx(X)
sf = SparseCoding(2, 7, max_iter=10)
for i, info in enumerate(sf.iter_fit(X)):
if i >= 10:
break
def test_mlp_fit():
X = np.random.standard_normal((10, 2))
Z = np.random.standard_normal((10, 1))
X, Z = theano_floatx(X, Z)
mlp = Mlp(2, [10], 1, ['tanh'], 'identity', 'squared', max_iter=10)
mlp.fit(X, Z)
def test_srnn_predict():
X = np.random.standard_normal((10, 5, 2)).astype(theano.config.floatX)
X, = theano_floatx(X)
rnn = SupervisedRnn(2, [10], 3, hidden_transfers=['tanh'], max_iter=10)
rnn.predict(X)
rnn = SupervisedRnn(2, [10], 3, hidden_transfers=['tanh'], skip_to_out=True, max_iter=10)
rnn.predict(X)
def test_mlp_iter_fit():
X = np.random.standard_normal((10, 2))
Z = np.random.standard_normal((10, 1))
X, Z = theano_floatx(X, Z)
mlp = Mlp(2, [10], 1, ['tanh'], 'identity', 'squared', max_iter=10)
for i, info in enumerate(mlp.iter_fit(X, Z)):
if i >= 10:
break
def test_mlp_fit_with_imp_weight():
X = np.random.standard_normal((10, 2))
Z = np.random.standard_normal((10, 1))
W = np.random.random((10, 1)) > 0.5
X, Z, W = theano_floatx(X, Z, W)
mlp = Mlp(2, [10], 1, ['tanh'], 'identity', 'squared', max_iter=10, imp_weight=True)
mlp.fit(X, Z, W)
def test_checkpoint_trainer():
# Make model and data for the test.
X = np.random.random((100, 10))
Z = np.random.random((100, 2))
X, Z = theano_floatx(X, Z)
data = {'train': (X, Z), 'val': (X, Z), 'test': (X, Z)}
optimizer = 'rmsprop', {'step_rate': 0.0001}
m = mlp.Mlp(10, [2],
2, ['tanh'],
'identity',
'squared',
max_iter=10,
optimizer=optimizer)
# Train the mdoel with a trainer for 2 epochs.
t = Trainer(m,
data,
stop=climin.stops.AfterNIterations(2),
pause=climin.stops.always)
t.val_key = 'val'
t.fit()
# Make a copy of the trainer.
t2 = copy.deepcopy(t)
t2.data = t.data
intermediate_pars = t2.model.parameters.data.copy()
intermediate_info = t2.current_info.copy()
# Train original for 2 more epochs.
t.stop = climin.stops.AfterNIterations(4)
t.fit()
# Check that the snapshot has not changed
assert np.all(t2.model.parameters.data == intermediate_pars)
final_pars = t.model.parameters.data.copy()
final_info = t.current_info.copy()
check_infos(intermediate_info, t2.current_info)
t2.stop = climin.stops.AfterNIterations(4)
t2.fit()
check_infos(final_info, t2.current_info)
assert np.allclose(final_pars, t2.model.parameters.data)
t_pickled = cPickle.dumps(t2)
t_unpickled = cPickle.loads(t_pickled)
t_unpickled.data = data
t.stop = climin.stops.AfterNIterations(4)
t_unpickled.fit()
assert np.allclose(final_pars,
t_unpickled.model.parameters.data,
atol=5.e-3)
def test_checkpoint_trainer():
# Make model and data for the test.
X = np.random.random((100, 10))
Z = np.random.random((100, 2))
X, Z = theano_floatx(X, Z)
data = {
'train': (X, Z),
'val': (X, Z),
'test': (X, Z)
}
optimizer = 'rmsprop', {'step_rate': 0.0001}
m = mlp.Mlp(10, [2], 2, ['tanh'], 'identity', 'squared', max_iter=10,
optimizer=optimizer)
# Train the mdoel with a trainer for 2 epochs.
t = Trainer(
m,
data,
stop=climin.stops.AfterNIterations(2),
pause=climin.stops.always)
t.val_key = 'val'
t.fit()
# Make a copy of the trainer.
t2 = copy.deepcopy(t)
t2.data = t.data
intermediate_pars = t2.model.parameters.data.copy()
intermediate_info = t2.current_info.copy()
# Train original for 2 more epochs.
t.stop = climin.stops.AfterNIterations(4)
t.fit()
# Check that the snapshot has not changed
assert np.all(t2.model.parameters.data == intermediate_pars)
final_pars = t.model.parameters.data.copy()
final_info = t.current_info.copy()
check_infos(intermediate_info, t2.current_info)
t2.stop = climin.stops.AfterNIterations(4)
t2.fit()
check_infos(final_info, t2.current_info)
assert np.allclose(final_pars, t2.model.parameters.data)
t_pickled = cPickle.dumps(t2)
t_unpickled = cPickle.loads(t_pickled)
t_unpickled.data = data
t.stop = climin.stops.AfterNIterations(4)
t_unpickled.fit()
assert np.allclose(
final_pars, t_unpickled.model.parameters.data, atol=5.e-3)
def test_srnn_lstm_fit():
X = np.random.standard_normal((13, 5, 4)).astype(theano.config.floatX)
Z = np.random.standard_normal((13, 5, 3)).astype(theano.config.floatX)
W = np.random.standard_normal((13, 5, 3)).astype(theano.config.floatX)
X, Z, W = theano_floatx(X, Z, W)
rnn = SupervisedRnn(4, [10], 3, hidden_transfers=['lstm'], max_iter=2)
rnn.fit(X, Z)
def test_rica_iter_fit():
X = np.random.standard_normal((10, 2))
X, = theano_floatx(X)
rica = Rica(2, 10, code_transfer='softabs',
hidden_transfer='identity', loss='squared',
c_ica=0.5, max_iter=10)
for i, info in enumerate(rica.iter_fit(X)):
if i >= 10:
break
def test_deep_fdvae():
X = np.random.random((2, 10))
X, = theano_floatx(X)
m = MyFDVAE(95, [20, 30],
4, [15, 25], ['rectifier'] * 2, ['rectifier'] * 2,
optimizer='rprop',
batch_size=None,
max_iter=3)