def test_config_fill_values():
vector_fill_values = ["1.0 0.0 1.04 10.49", "1 2 3 4 5" "0" "1.0" ""]
binary_fill_values = ["yes", "No", "1", "TRUE", 1]
for vector_fill_value, binary_fill_value in zip(vector_fill_values,
binary_fill_values):
config = {
"input_features": [
vector_feature(
preprocessing={"fill_value": vector_fill_value}),
],
"output_features":
[binary_feature(preprocessing={"fill_value": binary_fill_value})],
}
validate_config(config)
bad_vector_fill_values = ["one two three", "1,2,3", 0]
bad_binary_fill_values = ["one", 2, "maybe"]
for vector_fill_value, binary_fill_value in zip(bad_vector_fill_values,
bad_binary_fill_values):
config = {
"input_features": [
vector_feature(
preprocessing={"fill_value": vector_fill_value}),
],
"output_features":
[binary_feature(preprocessing={"fill_value": binary_fill_value})],
}
with pytest.raises(ValidationError):
validate_config(config)
def test_config_bad_preprocessing_param():
config = {
"input_features": [
sequence_feature(reduce_output="sum", encoder="fake"),
image_feature(
"/tmp/destination_folder",
preprocessing={
"in_memory": True,
"height": 12,
"width": 12,
"num_channels": 3,
"tokenizer": "space",
},
),
],
"output_features":
[category_feature(vocab_size=2, reduce_input="sum")],
"combiner": {
"type": "concat",
"output_size": 14
},
}
with pytest.raises(ValidationError, match=r"^'fake' is not one of .*"):
validate_config(config)
def test_config_input_output_features():
config = {
"input_features": [
category_feature(),
number_feature(),
],
"output_features": [binary_feature()],
}
validate_config(config)
def test_incorrect_output_features_config():
config = {
"input_features": [
number_feature(),
],
"output_features": [binary_feature(decoder="classifier")],
}
# Invalid decoder for binary output feature
with pytest.raises(ValidationError):
validate_config(config)
def test_config_bad_encoder_name():
config = {
"input_features":
[sequence_feature(reduce_output="sum", encoder="fake")],
"output_features":
[category_feature(vocab_size=2, reduce_input="sum")],
"combiner": {
"type": "concat",
"output_size": 14
},
}
with pytest.raises(ValidationError, match=r"^'fake' is not one of .*"):
validate_config(config)
def test_config_encoders():
for encoder in ENCODERS:
config = {
"input_features": [
sequence_feature(reduce_output="sum", encoder=encoder),
image_feature("/tmp/destination_folder"),
],
"output_features":
[category_feature(vocab_size=2, reduce_input="sum")],
"combiner": {
"type": "concat",
"output_size": 14
},
}
validate_config(config)
def test_validate_with_preprocessing_defaults():
config = {
"input_features": [
audio_feature(
"/tmp/destination_folder",
preprocessing=AudioFeatureMixin.preprocessing_defaults()),
bag_feature(
preprocessing=BagFeatureMixin.preprocessing_defaults()),
binary_feature(
preprocessing=BinaryFeatureMixin.preprocessing_defaults()),
category_feature(
preprocessing=CategoryFeatureMixin.preprocessing_defaults()),
date_feature(
preprocessing=DateFeatureMixin.preprocessing_defaults()),
h3_feature(preprocessing=H3FeatureMixin.preprocessing_defaults()),
image_feature(
"/tmp/destination_folder",
preprocessing=ImageFeatureMixin.preprocessing_defaults()),
number_feature(
preprocessing=NumberFeatureMixin.preprocessing_defaults()),
sequence_feature(
preprocessing=SequenceFeatureMixin.preprocessing_defaults()),
set_feature(
preprocessing=SetFeatureMixin.preprocessing_defaults()),
text_feature(
preprocessing=TextFeatureMixin.preprocessing_defaults()),
timeseries_feature(
preprocessing=TimeseriesFeatureMixin.preprocessing_defaults()),
vector_feature(
preprocessing=VectorFeatureMixin.preprocessing_defaults()),
],
"output_features": [{
"name": "target",
"type": "category"
}],
TRAINER: {
"decay": True,
"learning_rate": 0.001,
"validation_field": "target",
"validation_metric": "accuracy",
},
}
validate_config(config)
config = merge_with_defaults(config)
validate_config(config)
def test_incorrect_input_features_config():
config = {
"input_features": [
category_feature(preprocessing={"normalization": "zscore"}),
],
"output_features": [binary_feature()],
}
# Not a preprocessing param for category feature
with pytest.raises(ValidationError):
validate_config(config)
config = {
"input_features": [
text_feature(preprocessing={"padding_symbol": 0}),
],
"output_features": [binary_feature()],
}
# Incorrect type for padding_symbol preprocessing param
with pytest.raises(ValidationError):
validate_config(config)
config = {
"input_features": [
binary_feature(),
],
"output_features": [binary_feature()],
}
del config["input_features"][0]["type"]
# Incorrect type for padding_symbol preprocessing param
with pytest.raises(ValidationError):
validate_config(config)
def test_config_tabnet(eval_batch_size):
config = {
"input_features": [
category_feature(vocab_size=2, reduce_input="sum"),
number_feature(),
],
"output_features": [binary_feature(weight_regularization=None)],
"combiner": {
"type": "tabnet",
"size": 24,
"output_size": 26,
"sparsity": 0.000001,
"bn_virtual_divider": 32,
"bn_momentum": 0.4,
"num_steps": 5,
"relaxation_factor": 1.5,
"use_keras_batch_norm": False,
"bn_virtual_bs": 512,
},
TRAINER: {
"batch_size": 16384,
"eval_batch_size": eval_batch_size,
"epochs": 1000,
"early_stop": 20,
"learning_rate": 0.02,
"optimizer": {
"type": "adam"
},
"decay": True,
"decay_steps": 20000,
"decay_rate": 0.9,
"staircase": True,
"regularization_lambda": 1,
"regularization_type": "l2",
"validation_field": "label",
},
}
validate_config(config)
def test_config_features():
all_input_features = [
audio_feature("/tmp/destination_folder"),
bag_feature(),
binary_feature(),
category_feature(),
date_feature(),
h3_feature(),
image_feature("/tmp/destination_folder"),
number_feature(),
sequence_feature(),
set_feature(),
text_feature(),
timeseries_feature(),
vector_feature(),
]
all_output_features = [
binary_feature(),
category_feature(),
number_feature(),
sequence_feature(),
set_feature(),
text_feature(),
vector_feature(),
]
# validate config with all features
config = {
"input_features": all_input_features,
"output_features": all_output_features,
}
validate_config(config)
# make sure all defaults provided also registers as valid
config = merge_with_defaults(config)
validate_config(config)
# test various invalid output features
input_only_features = [
feature for feature in all_input_features
if feature["type"] not in output_type_registry.keys()
]
for input_feature in input_only_features:
config = {
"input_features": all_input_features,
"output_features": all_output_features + [input_feature],
}
dtype = input_feature["type"]
with pytest.raises(ValidationError,
match=rf"^'{dtype}' is not one of .*"):
validate_config(config)
def test_config_trainer_empty_null_and_default():
config = {
"input_features": [
category_feature(vocab_size=2, reduce_input="sum"),
number_feature(),
],
"output_features": [binary_feature(weight_regularization=None)],
"combiner": {
"type": "tabnet",
},
TRAINER: {},
}
validate_config(config)
config[TRAINER] = None
with pytest.raises(ValidationError):
validate_config(config)
config[TRAINER] = ECDTrainerConfig.Schema().dump({})
validate_config(config)
def test_optimizer_property_validation():
config = {
"input_features": [
category_feature(vocab_size=2, reduce_input="sum"),
number_feature(),
],
"output_features": [binary_feature(weight_regularization=None)],
"combiner": {
"type": "tabnet",
},
TRAINER: {},
}
validate_config(config)
# Test that an optimizer's property types are enforced:
config[TRAINER]["optimizer"] = {"type": "rmsprop"}
validate_config(config)
config[TRAINER]["optimizer"]["momentum"] = "invalid"
with pytest.raises(ValidationError):
validate_config(config)
# Test extra keys are excluded and defaults are loaded appropriately:
config[TRAINER]["optimizer"]["momentum"] = 10
config[TRAINER]["optimizer"]["extra_key"] = "invalid"
validate_config(config)
assert not hasattr(
ECDTrainerConfig.Schema().load(config[TRAINER]).optimizer, "extra_key")
# Test bad parameter range:
config[TRAINER]["optimizer"] = {"type": "rmsprop", "eps": -1}
with pytest.raises(ValidationError):
validate_config(config)
# Test config validation for tuple types:
config[TRAINER]["optimizer"] = {"type": "adam", "betas": (0.1, 0.1)}
validate_config(config)
def test_config_trainer_bad_optimizer():
config = {
"input_features": [
category_feature(vocab_size=2, reduce_input="sum"),
number_feature(),
],
"output_features": [binary_feature(weight_regularization=None)],
"combiner": {
"type": "tabnet",
},
TRAINER: {},
}
validate_config(config)
# Test manually set-to-null optimizer vs unspecified:
config[TRAINER]["optimizer"] = None
with pytest.raises(ValidationError):
validate_config(config)
assert ECDTrainerConfig.Schema().load({}).optimizer is not None
# Test all types in optimizer_registry supported:
for key in optimizer_registry.keys():
config[TRAINER]["optimizer"] = {"type": key}
validate_config(config)
# Test invalid optimizer type:
config[TRAINER]["optimizer"] = {"type": 0}
with pytest.raises(ValidationError):
validate_config(config)
config[TRAINER]["optimizer"] = {"type": {}}
with pytest.raises(ValidationError):
validate_config(config)
config[TRAINER]["optimizer"] = {"type": "invalid"}
with pytest.raises(ValidationError):
validate_config(config)
def test_clipper_property_validation():
config = {
"input_features": [
category_feature(vocab_size=2, reduce_input="sum"),
number_feature(),
],
"output_features": [binary_feature(weight_regularization=None)],
"combiner": {
"type": "tabnet",
},
TRAINER: {},
}
validate_config(config)
# Test null/empty clipper:
config[TRAINER]["gradient_clipping"] = None
validate_config(config)
config[TRAINER]["gradient_clipping"] = {}
validate_config(config)
assert (ECDTrainerConfig.Schema().load(
config[TRAINER]).gradient_clipping == ECDTrainerConfig.Schema().load(
{}).gradient_clipping)
# Test invalid clipper type:
config[TRAINER]["gradient_clipping"] = 0
with pytest.raises(ValidationError):
validate_config(config)
config[TRAINER]["gradient_clipping"] = "invalid"
with pytest.raises(ValidationError):
validate_config(config)
# Test that an optimizer's property types are enforced:
config[TRAINER]["gradient_clipping"] = {"clipglobalnorm": None}
validate_config(config)
config[TRAINER]["gradient_clipping"] = {"clipglobalnorm": 1}
validate_config(config)
config[TRAINER]["gradient_clipping"] = {"clipglobalnorm": "invalid"}
with pytest.raises(ValidationError):
validate_config(config)
# Test extra keys are excluded and defaults are loaded appropriately:
config[TRAINER]["gradient_clipping"] = {"clipnorm": 1}
config[TRAINER]["gradient_clipping"]["extra_key"] = "invalid"
validate_config(config)
assert not hasattr(
ECDTrainerConfig.Schema().load(config[TRAINER]).gradient_clipping,
"extra_key")
def test_config_bad_combiner_types_enums():
config = {
"input_features": [
category_feature(vocab_size=2, reduce_input="sum"),
number_feature(),
],
"output_features": [binary_feature(weight_regularization=None)],
"combiner": {
"type": "concat",
"weights_initializer": "zeros"
},
}
# config is valid at this point
validate_config(config)
# Test weights initializer:
config["combiner"]["weights_initializer"] = {"test": "fail"}
with pytest.raises(ValidationError, match=r"{'test': 'fail'} is not of*"):
validate_config(config)
config["combiner"]["weights_initializer"] = "fail"
with pytest.raises(ValidationError, match=r"'fail' is not of*"):
validate_config(config)
config["combiner"]["weights_initializer"] = {}
with pytest.raises(ValidationError, match=r"Failed validating 'type'"):
validate_config(config)
config["combiner"]["weights_initializer"] = {"type": "fail"}
with pytest.raises(ValidationError, match=r"'fail' is not one of*"):
validate_config(config)
config["combiner"]["weights_initializer"] = {"type": "normal", "stddev": 0}
validate_config(config)
# Test bias initializer:
del config["combiner"]["weights_initializer"]
config["combiner"]["bias_initializer"] = "kaiming_uniform"
validate_config(config)
config["combiner"]["bias_initializer"] = "fail"
with pytest.raises(ValidationError, match=r"'fail' is not of*"):
validate_config(config)
config["combiner"]["bias_initializer"] = {}
with pytest.raises(ValidationError, match=r"Failed validating 'type'"):
validate_config(config)
config["combiner"]["bias_initializer"] = {"type": "fail"}
with pytest.raises(ValidationError, match=r"'fail' is not one of*"):
validate_config(config)
config["combiner"]["bias_initializer"] = {"type": "zeros", "stddev": 0}
validate_config(config)
# Test norm:
del config["combiner"]["bias_initializer"]
config["combiner"]["norm"] = "batch"
validate_config(config)
config["combiner"]["norm"] = "fail"
with pytest.raises(ValidationError, match=r"'fail' is not one of*"):
validate_config(config)
# Test activation:
del config["combiner"]["norm"]
config["combiner"]["activation"] = "relu"
validate_config(config)
config["combiner"]["activation"] = 123
with pytest.raises(ValidationError, match=r"123 is not of type*"):
validate_config(config)
# Test reduce_output:
del config["combiner"]["activation"]
config2 = {**config}
config2["combiner"]["type"] = "tabtransformer"
config2["combiner"]["reduce_output"] = "sum"
validate_config(config)
config2["combiner"]["reduce_output"] = "fail"
with pytest.raises(ValidationError, match=r"'fail' is not one of*"):
validate_config(config2)
# Test reduce_output = None:
config2["combiner"]["reduce_output"] = None
validate_config(config2)
def test_config_bad_combiner():
config = {
"input_features": [
category_feature(vocab_size=2, reduce_input="sum"),
number_feature(),
],
"output_features": [binary_feature(weight_regularization=None)],
"combiner": {
"type": "tabnet",
},
}
# config is valid at this point
validate_config(config)
# combiner without type
del config["combiner"]["type"]
with pytest.raises(ValidationError, match=r"^'type' is a required .*"):
validate_config(config)
# bad combiner type
config["combiner"]["type"] = "fake"
with pytest.raises(ValidationError, match=r"^'fake' is not one of .*"):
validate_config(config)
# bad combiner format (list instead of dict)
config["combiner"] = [{"type": "tabnet"}]
with pytest.raises(ValidationError,
match=r"^\[\{'type': 'tabnet'\}\] is not of .*"):
validate_config(config)
# bad combiner parameter types
config["combiner"] = {
"type": "tabtransformer",
"num_layers": 10,
"dropout": False,
}
with pytest.raises(ValidationError, match=r"^False is not of type.*"):
validate_config(config)
# bad combiner parameter range
config["combiner"] = {
"type": "transformer",
"dropout": -1,
}
with pytest.raises(ValidationError, match=r"less than the minimum.*"):
validate_config(config)
