def test_train(self):
mcbm = MCBM(8, 4, 20)
parameters = mcbm._parameters()
mcbm.train(
randint(2, size=[mcbm.dim_in, 2000]),
randint(2, size=[mcbm.dim_out, 2000]),
parameters={
'verbosity': 0,
'max_iter': 0,
})
# parameters should not have changed
self.assertLess(max(abs(mcbm._parameters() - parameters)), 1e-20)
def callback(i, mcbm):
return
mcbm.train(
randint(2, size=[mcbm.dim_in, 10000]),
randint(2, size=[mcbm.dim_out, 10000]),
parameters={
'verbosity': 0,
'max_iter': 10,
'threshold': 0.,
'batch_size': 1999,
'callback': callback,
'cb_iter': 1,
})
def test_train(self):
mcbm = MCBM(8, 4, 20)
parameters = mcbm._parameters()
mcbm.train(
randint(2, size=[mcbm.dim_in, 2000]),
randint(2, size=[mcbm.dim_out, 2000]),
parameters={"verbosity": 0, "max_iter": 0},
)
# parameters should not have changed
self.assertLess(max(abs(mcbm._parameters() - parameters)), 1e-20)
def callback(i, mcbm):
return
mcbm.train(
randint(2, size=[mcbm.dim_in, 10000]),
randint(2, size=[mcbm.dim_out, 10000]),
parameters={
"verbosity": 0,
"max_iter": 10,
"threshold": 0.0,
"batch_size": 1999,
"callback": callback,
"cb_iter": 1,
},
)
def test_gradient(self):
mcbm = MCBM(5, 2, 10)
# choose random parameters
mcbm._set_parameters(randn(*mcbm._parameters().shape))
err = mcbm._check_gradient(
randint(2, size=[mcbm.dim_in, 1000]),
randint(2, size=[mcbm.dim_out, 1000]), 1e-5)
self.assertLess(err, 1e-8)
# test with regularization turned off
for param in ['priors', 'weights', 'features', 'pred', 'input_bias', 'output_bias']:
err = mcbm._check_gradient(
randint(2, size=[mcbm.dim_in, 1000]),
randint(2, size=[mcbm.dim_out, 1000]),
1e-5,
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',
})
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 = mcbm._check_gradient(
randint(2, size=[mcbm.dim_in, 1000]),
randint(2, size=[mcbm.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_features': {'strength': 0.5, 'norm': norm},
'regularize_predictors': {'strength': 0.5, 'norm': norm},
'regularize_weights': {'strength': 0.5, 'norm': norm},
})
self.assertLess(err, 1e-6)
self.assertFalse(any(isnan(
mcbm._parameter_gradient(
randint(2, size=[mcbm.dim_in, 1000]),
randint(2, size=[mcbm.dim_out, 1000]),
mcbm._parameters()))))
def test_gradient(self):
mcbm = MCBM(5, 2, 10)
# choose random parameters
mcbm._set_parameters(randn(*mcbm._parameters().shape))
err = mcbm._check_gradient(randint(2, size=[mcbm.dim_in, 1000]), randint(2, size=[mcbm.dim_out, 1000]), 1e-5)
self.assertLess(err, 1e-8)
# test with regularization turned off
for param in ["priors", "weights", "features", "pred", "input_bias", "output_bias"]:
err = mcbm._check_gradient(
randint(2, size=[mcbm.dim_in, 1000]),
randint(2, size=[mcbm.dim_out, 1000]),
1e-5,
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",
},
)
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 = mcbm._check_gradient(
randint(2, size=[mcbm.dim_in, 1000]),
randint(2, size=[mcbm.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_features": {"strength": 0.5, "norm": norm},
"regularize_predictors": {"strength": 0.5, "norm": norm},
"regularize_weights": {"strength": 0.5, "norm": norm},
},
)
self.assertLess(err, 1e-6)
self.assertFalse(
any(
isnan(
mcbm._parameter_gradient(
randint(2, size=[mcbm.dim_in, 1000]), randint(2, size=[mcbm.dim_out, 1000]), mcbm._parameters()
)
)
)
)
