def test_imputer_numeric_data(test_dir):
"""
Tests numeric encoder/featurizer only
"""
# Training data
N = 1000
x = np.random.uniform(-np.pi, np.pi, (N, ))
df = pd.DataFrame({'x': x, 'cos': np.cos(x), '*2': x * 2, '**2': x**2})
df_train, df_test = random_split(df, [.6, .4])
output_path = os.path.join(test_dir, "tmp", "real_data_experiment_numeric")
data_encoder_cols = [NumericalEncoder(['x'])]
data_cols = [NumericalFeaturizer('x', numeric_latent_dim=100)]
for target in ['*2', '**2', 'cos']:
label_encoder_cols = [NumericalEncoder([target], normalize=False)]
imputer = Imputer(data_featurizers=data_cols,
label_encoders=label_encoder_cols,
data_encoders=data_encoder_cols,
output_path=output_path)
imputer.fit(train_df=df_train,
learning_rate=1e-1,
num_epochs=100,
patience=5,
test_split=.3,
weight_decay=.0,
batch_size=128)
pred, metrics = imputer.transform_and_compute_metrics(df_test)
df_test['predictions_' + target] = pred[target].flatten()
print("Numerical metrics: {}".format(metrics[target]))
assert metrics[target] < 10
def test_imputer_load_read_exec_only_dir(tmpdir, data_frame):
import stat
# on shared build-fleet tests fail with converting tmpdir to string
tmpdir = str(tmpdir)
feature = 'feature'
label = 'label'
df = data_frame(feature, label, n_samples=100)
# fit and output model + metrics to tmpdir
imputer = Imputer(data_featurizers=[BowFeaturizer(feature)],
label_encoders=[CategoricalEncoder(label)],
data_encoders=[BowEncoder(feature)],
output_path=tmpdir)
imputer.fit(train_df=df, num_epochs=1)
# make tmpdir read/exec-only by owner/group/others
os.chmod(
tmpdir, stat.S_IEXEC | stat.S_IXGRP | stat.S_IXOTH | stat.S_IREAD
| stat.S_IRGRP | stat.S_IROTH)
try:
Imputer.load(tmpdir)
except AssertionError as e:
print(e)
pytest.fail(
'Loading imputer from read-only directory should not fail.')
def test_imputer_without_train_df(test_dir):
"""
Test asserting that imputer.fit fails without training data or training data in wrong format
"""
df_train = ['ffffffooooo']
data_encoder_cols = [BowEncoder('item_name')]
label_encoder_cols = [CategoricalEncoder('brand')]
data_cols = [BowFeaturizer('item_name')]
output_path = os.path.join(test_dir, "tmp", "real_data_experiment")
imputer = Imputer(
data_featurizers=data_cols,
label_encoders=label_encoder_cols,
data_encoders=data_encoder_cols,
output_path=output_path,
)
with pytest.raises(
ValueError,
message="Need a non-empty DataFrame for fitting Imputer model"):
imputer.fit(train_df=df_train)
with pytest.raises(
ValueError,
message="Need a non-empty DataFrame for fitting Imputer model"):
imputer.fit(train_df=None)
def test_imputer_without_test_set_random_split():
"""
Test asserting that the random split is working internally
by calling imputer.fit only with a training set.
"""
feature_col = "string_feature"
label_col = "label"
n_samples = 5000
num_labels = 3
seq_len = 20
vocab_size = int(2 ** 10)
# generate some random data
df_train = generate_string_data_frame(feature_col=feature_col,
label_col=label_col,
vocab_size=vocab_size,
num_labels=num_labels,
num_words=seq_len,
n_samples=n_samples)
num_epochs = 1
batch_size = 64
learning_rate = 1e-3
data_encoder_cols = [
BowEncoder(feature_col, max_tokens=vocab_size)
]
label_encoder_cols = [CategoricalEncoder(label_col, max_tokens=num_labels)]
data_cols = [
BowFeaturizer(feature_col, vocab_size=vocab_size)
]
output_path = os.path.join(dir_path, "resources", "tmp", "real_data_experiment")
imputer = Imputer(
data_featurizers=data_cols,
label_encoders=label_encoder_cols,
data_encoders=data_encoder_cols,
output_path=output_path
)
try:
imputer.fit(
train_df=df_train,
learning_rate=learning_rate,
num_epochs=num_epochs,
batch_size=batch_size
)
except TypeError:
pytest.fail("Didn't expect a TypeError exception with missing test data")
shutil.rmtree(output_path)
def test_imputer_load_with_invalid_context(tmpdir, data_frame):
# on shared build-fleet tests fail with converting tmpdir to string
tmpdir = str(tmpdir)
feature = 'feature'
label = 'label'
df = data_frame(feature, label, n_samples=100)
# fit and output model + metrics to tmpdir
imputer = Imputer(data_featurizers=[BowFeaturizer(feature)],
label_encoders=[CategoricalEncoder(label)],
data_encoders=[BowEncoder(feature)],
output_path=tmpdir)
imputer.fit(train_df=df, num_epochs=1)
imputer.ctx = None
imputer.save()
imputer_deser = Imputer.load(tmpdir)
_ = imputer_deser.predict(df)
def test_imputer_fit_fail_non_writable_output_dir(tmpdir, data_frame):
import stat
# on shared build-fleet tests fail with converting tmpdir to string
tmpdir = str(tmpdir)
feature = 'feature'
label = 'label'
df = data_frame(feature, label, n_samples=100)
# fit and output model + metrics to tmpdir
imputer = Imputer(data_featurizers=[BowFeaturizer(feature)],
label_encoders=[CategoricalEncoder(label)],
data_encoders=[BowEncoder(feature)],
output_path=tmpdir)
# make tmpdir read/exec-only by owner/group/others
os.chmod(
tmpdir, stat.S_IEXEC | stat.S_IXGRP | stat.S_IXOTH | stat.S_IREAD
| stat.S_IRGRP | stat.S_IROTH)
# fail if imputer.fit does not raise an AssertionError
with pytest.raises(AssertionError) as e:
imputer.fit(df, num_epochs=1)
def test_fit_resumes(test_dir, data_frame):
feature_col, label_col = "feature", "label"
df = data_frame(feature_col=feature_col, label_col=label_col)
imputer = Imputer(
data_encoders=[TfIdfEncoder([feature_col])],
data_featurizers=[
datawig.mxnet_input_symbols.BowFeaturizer(feature_col)
],
label_encoders=[CategoricalEncoder(label_col)],
output_path=test_dir)
assert imputer.module is None
imputer.fit(df, num_epochs=20)
first_fit_module = imputer.module
imputer.fit(df, num_epochs=20)
second_fit_module = imputer.module
assert first_fit_module == second_fit_module
def test_imputer_duplicate_encoder_output_columns(test_dir, data_frame):
"""
Tests Imputer with sequential, bag-of-words and categorical variables as inputs
this could be run as part of integration test suite.
"""
feature_col = "string_feature"
categorical_col = "categorical_feature"
label_col = "label"
n_samples = 1000
num_labels = 10
seq_len = 100
vocab_size = int(2**10)
latent_dim = 30
embed_dim = 30
# generate some random data
random_data = data_frame(feature_col=feature_col,
label_col=label_col,
vocab_size=vocab_size,
num_labels=num_labels,
num_words=seq_len,
n_samples=n_samples)
# we use a the label prefixes as a dummy categorical input variable
random_data[categorical_col] = random_data[label_col].apply(
lambda x: x[:2])
df_train, df_test, df_val = random_split(random_data, [.8, .1, .1])
data_encoder_cols = [
BowEncoder(feature_col, feature_col, max_tokens=vocab_size),
SequentialEncoder(feature_col,
feature_col,
max_tokens=vocab_size,
seq_len=seq_len),
CategoricalEncoder(categorical_col, max_tokens=num_labels)
]
label_encoder_cols = [CategoricalEncoder(label_col, max_tokens=num_labels)]
data_cols = [
BowFeaturizer(feature_col, vocab_size=vocab_size),
LSTMFeaturizer(field_name=feature_col,
seq_len=seq_len,
latent_dim=latent_dim,
num_hidden=30,
embed_dim=embed_dim,
num_layers=2,
vocab_size=num_labels),
EmbeddingFeaturizer(field_name=categorical_col,
embed_dim=embed_dim,
vocab_size=num_labels)
]
output_path = os.path.join(test_dir, "tmp",
"imputer_experiment_synthetic_data")
num_epochs = 20
batch_size = 16
learning_rate = 1e-3
with pytest.raises(ValueError) as e:
imputer = Imputer(data_featurizers=data_cols,
label_encoders=label_encoder_cols,
data_encoders=data_encoder_cols,
output_path=output_path)
imputer.fit(train_df=df_train,
test_df=df_val,
learning_rate=learning_rate,
num_epochs=num_epochs,
batch_size=batch_size)
