def log_training_data(self,
train_features,
train_targets,
overwrite=False):
"""
Associate training data with this model reference.
.. versionchanged:: 0.14.4
Instead of uploading the data itself as a CSV artifact ``'train_data'``, this method now
generates a histogram for internal use by our deployment data monitoring system.
.. deprecated:: 0.18.0
This method is no longer supported. Please see our documentation
for information about our platform's data monitoring features.
Parameters
----------
train_features : pd.DataFrame
pandas DataFrame representing features of the training data.
train_targets : pd.DataFrame or pd.Series
pandas DataFrame representing targets of the training data.
overwrite : bool, default False
Whether to allow overwriting existing training data.
"""
warnings.warn(
"This method is no longer supported. Please see our documentation"
" for information about our platform's data monitoring features",
category=FutureWarning,
)
if train_features.__class__.__name__ != "DataFrame":
raise TypeError(
"`train_features` must be a pandas DataFrame, not {}".format(
type(train_features)))
if train_targets.__class__.__name__ == "Series":
train_targets = train_targets.to_frame()
elif train_targets.__class__.__name__ != "DataFrame":
raise TypeError(
"`train_targets` must be a pandas DataFrame or Series, not {}".
format(type(train_targets)))
# check for overlapping column names
common_column_names = set(train_features.columns) & set(
train_targets.columns)
if common_column_names:
raise ValueError(
"`train_features` and `train_targets` combined have overlapping column names;"
" please ensure column names are unique")
train_df = train_features.join(train_targets)
histograms = _histogram_utils.calculate_histograms(train_df)
response = _utils.make_request("PUT",
self._histogram_endpoint,
self._conn,
json=histograms)
_utils.raise_for_http_error(response)
def test_integration(self, experiment_run):
np = pytest.importorskip("numpy")
pd = pytest.importorskip("pandas")
binary_col_name = 'binary col'
discrete_col_name = 'discrete col'
float_col_name = 'float col'
df = pd.concat(
objs=[
pd.Series([True] * 10 + [False] * 20, name=binary_col_name),
pd.Series([0] * 5 + [1] * 10 + [2] * 15,
name=discrete_col_name),
pd.Series(range(30), name=float_col_name),
],
axis='columns',
)
histograms = _histogram_utils.calculate_histograms(df)
experiment_run.log_training_data(
df[[binary_col_name, discrete_col_name]], df[float_col_name])
endpoint = "{}://{}/api/v1/monitoring/data/references/{}".format(
experiment_run._conn.scheme,
experiment_run._conn.socket,
experiment_run.id,
)
response = _utils.make_request("GET", endpoint, experiment_run._conn)
_utils.raise_for_http_error(response)
retrieved_histograms = response.json()
# features match
features = histograms['features']
retrieved_features = retrieved_histograms['features']
assert set(features.keys()) == set(retrieved_features.keys())
# binary matches
binary_hist = histograms['features'][binary_col_name]['histogram'][
'binary']
retrieved_binary_hist = retrieved_histograms['features'][
binary_col_name]['histogram']['binary']
assert binary_hist['count'] == retrieved_binary_hist['count']
# discrete matches
discrete_hist = histograms['features'][discrete_col_name]['histogram'][
'discrete']
retrieved_discrete_hist = retrieved_histograms['features'][
discrete_col_name]['histogram']['discrete']
assert discrete_hist['bucket_values'] == retrieved_discrete_hist[
'bucket_values']
assert discrete_hist['count'] == retrieved_discrete_hist['count']
# float matches
float_hist = histograms['features'][float_col_name]['histogram'][
'float']
retrieved_float_hist = retrieved_histograms['features'][
float_col_name]['histogram']['float']
assert all(
np.isclose(float_hist['bucket_limits'],
retrieved_float_hist['bucket_limits']))
assert float_hist['count'] == retrieved_float_hist['count']
def test_discrete(self):
np = pytest.importorskip("numpy")
pd = pytest.importorskip("pandas")
num_rows = 90
df = pd.concat(
objs=[
pd.Series(np.random.randint(6, 12, size=num_rows), name="A"),
pd.Series(np.random.randint(-12, -6, size=num_rows), name="B"),
pd.Series(np.random.randint(-3, 3, size=num_rows), name="C"),
],
axis='columns',
)
histograms = _histogram_utils.calculate_histograms(df)
assert all(histogram['type'] == "discrete"
for histogram in histograms['features'].values())
self.assert_histograms_match_dataframe(histograms, df)
def test_float(self):
np = pytest.importorskip("numpy")
pd = pytest.importorskip("pandas")
num_rows = 90
df = pd.concat(
objs=[
pd.Series(np.random.normal(loc=9, size=num_rows), name="A"),
pd.Series(np.random.normal(scale=12, size=num_rows), name="B"),
pd.Series(np.random.normal(loc=-3, scale=6, size=num_rows),
name="C"),
],
axis='columns',
)
histograms = _histogram_utils.calculate_histograms(df)
assert all(histogram['type'] == "float"
for histogram in histograms['features'].values())
self.assert_histograms_match_dataframe(histograms, df)
def test_binary(self):
np = pytest.importorskip("numpy")
pd = pytest.importorskip("pandas")
num_rows = 90
df = pd.concat(
objs=[
pd.Series(np.random.random(size=num_rows).round().astype(bool),
name="A"),
pd.Series(np.random.random(size=num_rows).round().astype(bool),
name="B"),
pd.Series(np.random.random(size=num_rows).round().astype(bool),
name="C"),
],
axis='columns',
)
histograms = _histogram_utils.calculate_histograms(df)
assert all(histogram['type'] == "binary"
for histogram in histograms['features'].values())
self.assert_histograms_match_dataframe(histograms, df)
