def predict_with_backend(tmpdir,
config,
data_csv_path,
backend,
patch_args=None):
with init_backend(backend):
if backend == "ray":
backend = RAY_BACKEND_CONFIG
backend["processor"]["type"] = "dask"
ludwig_model = LudwigModel(config, backend=backend)
_, _, output_directory = ludwig_model.train(
dataset=data_csv_path,
output_directory=os.path.join(tmpdir, "output"),
)
# Check that metadata JSON saves and loads correctly
ludwig_model = LudwigModel.load(os.path.join(output_directory,
"model"))
if patch_args is not None:
with mock.patch(*patch_args):
preds_df, _ = ludwig_model.predict(dataset=data_csv_path)
else:
preds_df, _ = ludwig_model.predict(dataset=data_csv_path)
return preds_df, ludwig_model
def test_empty_split_error(backend, tmpdir):
"""Tests that an error is raised if one or more of the splits is empty after preprocessing."""
data_csv_path = os.path.join(tmpdir, "data.csv")
out_feat = binary_feature()
input_features = [number_feature()]
output_features = [out_feat]
config = {
"input_features": input_features,
"output_features": output_features
}
training_data_csv_path = generate_data(input_features, output_features,
data_csv_path)
df = pd.read_csv(training_data_csv_path)
# Convert all the output features rows to null. Because the default missing value strategy is to drop empty output
# rows, this will result in the dataset being empty after preprocessing.
df[out_feat[COLUMN]] = None
with init_backend(backend):
ludwig_model = LudwigModel(config, backend=backend)
with pytest.raises(ValueError,
match="Dataset is empty following preprocessing"):
ludwig_model.preprocess(dataset=df)
def test_sample_ratio(backend, tmpdir):
num_examples = 100
sample_ratio = 0.25
input_features = [sequence_feature(reduce_output="sum")]
output_features = [category_feature(vocab_size=5, reduce_input="sum")]
data_csv = generate_data(input_features,
output_features,
os.path.join(tmpdir, "dataset.csv"),
num_examples=num_examples)
config = {
"input_features": input_features,
"output_features": output_features,
"trainer": {
"epochs": 2,
},
"preprocessing": {
"sample_ratio": sample_ratio
},
}
with init_backend(backend):
model = LudwigModel(config, backend=backend)
train_set, val_set, test_set, _ = model.preprocess(
data_csv,
skip_save_processed_input=True,
)
sample_size = num_examples * sample_ratio
count = len(train_set) + len(val_set) + len(test_set)
assert sample_size == count
def test_dask_known_divisions(feature_fn, csv_filename, tmpdir):
import dask.dataframe as dd
input_features = [feature_fn(os.path.join(tmpdir, "generated_output"))]
output_features = [category_feature(vocab_size=5, reduce_input="sum")]
# num_examples=100 and npartitions=2 to ensure the test is not flaky, by having non-empty post-split datasets.
data_csv = generate_data(input_features,
output_features,
os.path.join(tmpdir, csv_filename),
num_examples=100)
data_df = dd.from_pandas(pd.read_csv(data_csv), npartitions=2)
assert data_df.known_divisions
config = {
"input_features": input_features,
"output_features": output_features,
"trainer": {
"epochs": 2,
},
}
backend = "ray"
with init_backend(backend):
model = LudwigModel(config, backend=backend)
train_set, val_set, test_set, _ = model.preprocess(
data_df,
skip_save_processed_input=False,
)
def test_missing_values_fill_with_mean(backend, csv_filename, tmpdir):
data_csv_path = os.path.join(tmpdir, csv_filename)
kwargs = {PREPROCESSING: {"missing_value_strategy": FILL_WITH_MEAN}}
input_features = [
number_feature(**kwargs),
binary_feature(),
category_feature(vocab_size=3),
]
output_features = [binary_feature()]
training_data_csv_path = generate_data(input_features, output_features, data_csv_path)
config = {"input_features": input_features, "output_features": output_features, TRAINER: {"epochs": 2}}
with init_backend(backend):
# run preprocessing
ludwig_model = LudwigModel(config, backend=backend)
ludwig_model.preprocess(dataset=training_data_csv_path)
def test_with_split(backend, csv_filename, tmpdir):
num_examples = NUM_EXAMPLES
train_set_size = int(num_examples * 0.8)
val_set_size = int(num_examples * 0.1)
test_set_size = int(num_examples * 0.1)
input_features = [sequence_feature(reduce_output="sum")]
output_features = [category_feature(vocab_size=5, reduce_input="sum")]
data_csv = generate_data(input_features,
output_features,
os.path.join(tmpdir, csv_filename),
num_examples=num_examples)
data_df = pd.read_csv(data_csv)
data_df["split"] = [0] * train_set_size + [1] * val_set_size + [
2
] * test_set_size
data_df.to_csv(data_csv, index=False)
config = {
"input_features": input_features,
"output_features": output_features,
"trainer": {
"epochs": 2,
},
"preprocessing": {
"split": {
"type": "fixed"
}
},
}
with init_backend(backend):
model = LudwigModel(config, backend=backend)
train_set, val_set, test_set, _ = model.preprocess(
data_csv,
skip_save_processed_input=False,
)
assert len(train_set) == train_set_size
assert len(val_set) == val_set_size
assert len(test_set) == test_set_size