def test_save_and_load_weights(model_save_path):
arch = resnet18(2, use_pretrained=False)
model = cnn.CNN("resnet18", classes=["a", "b"], sample_duration=5.0)
model.save_weights(model_save_path)
model1 = cnn.CNN(arch, classes=["a", "b"], sample_duration=5.0)
model1.load_weights(model_save_path)
assert np.array_equal(
model.network.state_dict()["conv1.weight"].numpy(),
model1.network.state_dict()["conv1.weight"].numpy(),
)
def test_single_target_prediction(test_df):
model = cnn.CNN("resnet18", classes=[0, 1], sample_duration=5.0)
model.single_target = True
scores, preds, _ = model.predict(test_df, binary_preds="single_target")
assert len(scores) == 2
assert len(preds) == 2
def test_predict_splitting_short_file(short_file_df):
model = cnn.CNN("resnet18", classes=[0, 1], sample_duration=5.0)
with warnings.catch_warnings(record=True) as w:
warnings.simplefilter("always")
scores, _, _ = model.predict(short_file_df)
assert len(scores) == 0
assert "prediction_dataset" in str(w[0].message)
def test_split_resnet_feat_clf(train_df):
model = cnn.CNN("resnet18", classes=[0, 1], sample_duration=2)
cnn.separate_resnet_feat_clf(model)
assert "feature" in model.optimizer_params
model.optimizer_params["feature"]["lr"] = 0.1
model.train(train_df, epochs=0, save_path="tests/models")
shutil.rmtree("tests/models/")
def test_predict_without_splitting(test_df):
model = cnn.CNN("resnet18", classes=[0, 1], sample_duration=5.0)
scores, preds, _ = model.predict(test_df,
split_files_into_clips=False,
binary_preds="multi_target",
threshold=0)
assert len(scores) == len(test_df)
assert len(preds) == len(test_df)
def test_multi_target_prediction(train_df, test_df):
model = cnn.CNN("resnet18", classes=[0, 1], sample_duration=5.0)
scores, preds, _ = model.predict(test_df,
binary_preds="multi_target",
threshold=0.1)
assert len(scores) == 2
assert len(preds) == 2
def test_predict_wrong_input_error(test_df):
"""cannot pass a preprocessor or dataset to predict. only file paths as list or df"""
model = cnn.CNN("resnet18", classes=[0, 1], sample_duration=5.0)
pre = SpectrogramPreprocessor(2.0)
with pytest.raises(AssertionError):
model.predict(pre)
with pytest.raises(AssertionError):
ds = AudioFileDataset(test_df, pre)
model.predict(ds)
def test_prediction_overlap(test_df):
model = cnn.CNN("resnet18", classes=[0, 1], sample_duration=5.0)
model.single_target = True
scores, preds, _ = model.predict(test_df,
binary_preds="single_target",
overlap_fraction=0.5)
assert len(scores) == 3
assert len(preds) == 3
def test_prediction_warns_different_classes(train_df):
model = cnn.CNN("resnet18", classes=["a", "b"], sample_duration=5.0)
with warnings.catch_warnings(record=True) as w:
warnings.simplefilter("always")
# raises warning bc test_df columns != model.classes
model.predict(train_df)
all_warnings = ""
for wi in w:
all_warnings += str(wi.message)
assert "classes" in all_warnings
def test_train_multi_target(train_df):
model = cnn.CNN("resnet18", classes=[0, 1], sample_duration=5.0)
model.train(
train_df,
train_df,
save_path="tests/models",
epochs=1,
batch_size=2,
save_interval=10,
num_workers=0,
)
shutil.rmtree("tests/models/")
def test_save_and_load_model(model_save_path):
arch = alexnet(2, use_pretrained=False)
classes = [0, 1]
cnn.CNN(arch, classes, 1.0).save(model_save_path)
m = cnn.load_model(model_save_path)
assert m.classes == classes
assert type(m) == cnn.CNN
cnn.InceptionV3(classes, 1.0, use_pretrained=False).save(model_save_path)
m = cnn.load_model(model_save_path)
assert m.classes == classes
assert type(m) == cnn.InceptionV3
def test_train_one_class(train_df):
model = cnn.CNN("resnet18", classes=[0], sample_duration=5.0)
model.single_target = True
model.train(
train_df[[0]],
train_df[[0]],
save_path="tests/models",
epochs=1,
batch_size=2,
save_interval=10,
num_workers=0,
)
shutil.rmtree("tests/models/")
def test_train_predict_architecture(train_df):
"""test passing a specific architecture to PytorchModel"""
arch = alexnet(2, use_pretrained=False)
model = cnn.CNN(arch, [0, 1], sample_duration=2)
model.train(
train_df,
train_df,
save_path="tests/models/",
epochs=1,
batch_size=2,
save_interval=10,
num_workers=0,
)
model.predict(train_df, num_workers=0)
shutil.rmtree("tests/models/")
def test_init_with_str():
model = cnn.CNN("resnet18", classes=[0, 1], sample_duration=5.0)
def test_train_no_validation(train_df):
model = cnn.CNN("resnet18", classes=[0, 1], sample_duration=2)
model.train(train_df, save_path="tests/models")
shutil.rmtree("tests/models/")
def test_predict_missing_file_is_unsafe_sample(missing_file_df):
model = cnn.CNN("resnet18", classes=[0, 1], sample_duration=5.0)
scores, _, unsafe_samples = model.predict(missing_file_df, threshold=0.1)
assert len(scores) == 0
assert len(unsafe_samples) == 1
def test_eval(train_df):
model = cnn.CNN("resnet18", classes=[0, 1], sample_duration=2)
scores, _, _ = model.predict(train_df, split_files_into_clips=False)
model.eval(train_df.values, scores.values)
def test_prediction_returns_consistent_values(train_df):
model = cnn.CNN("resnet18", classes=["a", "b"], sample_duration=5.0)
a, _, _ = model.predict(train_df)
b, _, _ = model.predict(train_df)
assert np.allclose(a.values, b.values, 1e-6)