def test_train(self):
stm = STM(8, 4, 4, 10)
parameters = stm._parameters()
stm.train(
randint(2, size=[stm.dim_in, 2000]),
randint(2, size=[stm.dim_out, 2000]),
parameters={
'verbosity': 0,
'max_iter': 0,
})
# parameters should not have changed
self.assertLess(max(abs(stm._parameters() - parameters)), 1e-20)
def callback(i, stm):
callback.counter += 1
return
callback.counter = 0
max_iter = 10
stm.train(
randint(2, size=[stm.dim_in, 10000]),
randint(2, size=[stm.dim_out, 10000]),
parameters={
'verbosity': 0,
'max_iter': max_iter,
'threshold': 0.,
'batch_size': 1999,
'callback': callback,
'cb_iter': 2,
})
self.assertEqual(callback.counter, max_iter / 2)
# test zero-dimensional nonlinear inputs
stm = STM(0, 5, 5)
glm = GLM(stm.dim_in_linear, LogisticFunction, Bernoulli)
glm.weights = randn(*glm.weights.shape)
input = randn(stm.dim_in_linear, 10000)
output = glm.sample(input)
stm.train(input, output, parameters={'max_iter': 20})
# STM should be able to learn GLM behavior
self.assertAlmostEqual(glm.evaluate(input, output), stm.evaluate(input, output), 1)
# test zero-dimensional inputs
stm = STM(0, 0, 10)
input = empty([0, 10000])
output = rand(1, 10000) < 0.35
stm.train(input, output)
self.assertLess(abs(mean(stm.sample(input)) - mean(output)), 0.1)
def test_gradient(self):
stm = STM(5, 2, 10)
stm.sharpness = 1.5
# choose random parameters
stm._set_parameters(randn(*stm._parameters().shape) / 100.)
err = stm._check_gradient(
randn(stm.dim_in, 1000),
randint(2, size=[stm.dim_out, 1000]),
1e-5,
parameters={'train_sharpness': True})
self.assertLess(err, 1e-8)
# test with regularization turned off
for param in ['biases', 'weights', 'features', 'pred', 'linear_predictor', 'sharpness']:
err = stm._check_gradient(
randn(stm.dim_in, 1000),
randint(2, size=[stm.dim_out, 1000]),
1e-6,
parameters={
'train_biases': param == 'biases',
'train_weights': param == 'weights',
'train_features': param == 'features',
'train_predictors': param == 'pred',
'train_linear_predictor': param == 'linear_predictor',
'train_sharpness': param == 'sharpness',
})
self.assertLess(err, 1e-7)
# test with regularization turned on
for norm in ['L1', 'L2']:
for param in ['priors', 'weights', 'features', 'pred', 'input_bias', 'output_bias']:
err = stm._check_gradient(
randint(2, size=[stm.dim_in, 1000]),
randint(2, size=[stm.dim_out, 1000]),
1e-7,
parameters={
'train_prior': param == 'priors',
'train_weights': param == 'weights',
'train_features': param == 'features',
'train_predictors': param == 'pred',
'train_input_bias': param == 'input_bias',
'train_output_bias': param == 'output_bias',
'regularize_biases': {'strength': 0.6, 'norm': norm},
'regularize_features': {'strength': 0.6, 'norm': norm},
'regularize_predictors': {'strength': 0.6, 'norm': norm},
'regularize_weights': {'strength': 0.6, 'norm': norm},
})
self.assertLess(err, 1e-6)
self.assertFalse(any(isnan(
stm._parameter_gradient(
randint(2, size=[stm.dim_in, 1000]),
randint(2, size=[stm.dim_out, 1000]),
stm._parameters()))))
def test_poisson(self): stm = STM(5, 5, 3, 10, ExponentialFunction, Poisson) # choose random parameters stm._set_parameters(randn(*stm._parameters().shape) / 100.) err = stm._check_gradient( randn(stm.dim_in, 1000), randint(2, size=[stm.dim_out, 1000]), 1e-5) self.assertLess(err, 1e-8)
