def test_produce():
mlp = MlpClassifierPrimitive(
hyperparams=mlp_hp(mlp_hp.defaults(),
weights_filepath='/scratch_dir/model_weights.pth',
all_confidences=False))
mlp.set_params(params=mlp_params)
preds = mlp.produce(inputs=features).value
assert preds.shape == (features.shape[0], 2)
assert (preds.columns == ['target', 'confidence']).all()
def test_produce_explanations_all_classes():
mlp = MlpClassifierPrimitive(
hyperparams=mlp_hp(mlp_hp.defaults(),
weights_filepath='/scratch_dir/model_weights.pth',
explain_all_classes=True))
mlp.set_params(params=mlp_params)
explanations = mlp.produce_explanations(inputs=features).value
assert explanations.shape == (features.shape[0], mlp._nclasses)
assert np.array(explanations.iloc[0, 0]).shape == (120, 120)
def test_fit():
mlp = MlpClassifierPrimitive(
hyperparams=mlp_hp(mlp_hp.defaults(),
epochs=1,
weights_filepath='/scratch_dir/model_weights.pth'))
mlp.set_training_data(inputs=features, outputs=labels)
mlp.fit()
assert mlp._clf_model[-1].weight.shape[0] == mlp._nclasses
global mlp_params
mlp_params = mlp.get_params()
def test_produce_all_confidences():
mlp = MlpClassifierPrimitive(hyperparams=mlp_hp(
mlp_hp.defaults(),
weights_filepath='/scratch_dir/model_weights.pth',
))
mlp.set_params(params=mlp_params)
preds = mlp.produce(inputs=features).value
nc = mlp._nclasses
assert preds.shape == (features.shape[0] * nc, 2)
if nc > 2:
for i in range(0, features.shape[0]):
assert round(preds['confidence'][i * nc:(i + 1) * nc].sum()) == 1
assert (preds.columns == ['target', 'confidence']).all()