def test_multiple_hparams(self):
opt_dict = {
'hp-beta1':
optim.GradientDescent(learning_rate=1
).create(np.array([0.0, 0.1, 0.2, 0.9])),
'hp-learning_rate':
optim.GradientDescent(learning_rate=1
).create(np.array([0.0, 0.1, 0.2, 0.9])),
}
gv_dict = {
'beta1': {
'learning_rate_scalar': 1,
'activation_fn': 'linear',
'activation_ceiling': None,
'activation_floor': None,
'clip_min': None,
'clip_max': None,
},
'learning_rate': {
'learning_rate_scalar': 1,
'activation_fn': 'linear',
'activation_ceiling': None,
'activation_floor': None,
'clip_min': None,
'clip_max': None,
}
}
out = guided_parameters.get_activated_hparams(opt_dict, gv_dict)
self.assertListEqual(list(out['hp-beta1']), [0.0, 0.1, 0.2, 0.9])
self.assertListEqual(list(out['hp-learning_rate']), [0.0, 0.1, 0.2, 0.9])
def test_empty(self):
opt_dict = {}
gv_dict = {}
out = guided_parameters.get_activated_hparams(opt_dict, gv_dict)
self.assertEqual(
out, {
'hp-beta1': None,
'hp-decay_rate': None,
'hp-eps': None,
'hp-learning_rate': None,
'hp-weight_decay': None,
'hp-label_smoothing': None,
})
def test_ignores_model(self):
opt_dict = {
'model':
optim.GradientDescent(learning_rate=1
).create(np.array([0.0, 0.1, 0.2, 0.9])),
}
gv_dict = {}
out = guided_parameters.get_activated_hparams(opt_dict, gv_dict)
self.assertEqual(
out, {
'hp-beta1': None,
'hp-decay_rate': None,
'hp-eps': None,
'hp-learning_rate': None,
'hp-weight_decay': None,
'hp-label_smoothing': None,
})
