Skip to content

Commit

Permalink
Update dataset properties (#249)
Browse files Browse the repository at this point in the history 
* update dataset init properties

* Update column info notebook

* add tests for dataset label

* fix PR comments

* fix notebooks

Co-authored-by: Matan Perlmutter <matan@deepchecks.com>
  • Loading branch information
@matanper
matanper and Matan Perlmutter committed Dec 15, 2021
1 parent ba4931c commit 41053da
Show file tree
Hide file tree
Showing 56 changed files with 1,662 additions and 1,693 deletions.
165 changes: 77 additions & 88 deletions deepchecks/base/dataset.py
View file
Original file line number Diff line number Diff line change
Expand Up @@ -50,8 +50,8 @@ class Dataset:
"""

_features: t.List[Hashable]
_label: t.Optional[Hashable]
_use_index: bool
_label_name: t.Optional[Hashable]
_use_default_index: bool
_index_name: t.Optional[Hashable]
_date_name: t.Optional[Hashable]
cat_features: t.List[Hashable]
Expand All @@ -62,12 +62,13 @@ class Dataset:
def __init__(
self,
df: pd.DataFrame,
label: pd.Series = None,
features: t.Optional[t.Sequence[Hashable]] = None,
cat_features: t.Optional[t.Sequence[Hashable]] = None,
label: t.Optional[Hashable] = None,
use_index: bool = False,
index: t.Optional[Hashable] = None,
date: t.Optional[Hashable] = None,
label_name: t.Optional[Hashable] = None,
use_default_index: bool = False,
index_name: t.Optional[Hashable] = None,
date_name: t.Optional[Hashable] = None,
date_args: t.Optional[t.Dict] = None,
convert_date_: bool = True,
max_categorical_ratio: float = 0.01,
Expand All @@ -78,19 +79,22 @@ def __init__(
Args:
df (pandas.DataFrame):
A pandas DataFrame containing data relevant for the training or validating of a ML models
A pandas DataFrame containing data relevant for the training or validating of a ML models.
label (pandas.Series)
A pandas series containing data of the labels.
features (Optional[Sequence[Hashable]]):
List of names for the feature columns in the DataFrame.
cat_features (Optional[Sequence[Hashable]]):
List of names for the categorical features in the DataFrame. In order to disable categorical
List of names for the categorical features in the DataFrame. In order to disable categorical.
features inference, pass cat_features=[]
label (Optional[Hashable]):
label_name (Optional[Hashable]):
Name of the label column in the DataFrame.
use_index (bool, default False):
Whether to use the dataframe index as the index column, for index related checks.
index (Optional[Hashable]):
Name of the index column in the DataFrame.
date (Optional[Hashable]):
use_default_index (bool, default False):
Whether to use the dataframe index as the index column, for index related checks. can't be used
together with `index_name`
index_name (Optional[Hashable]):
Name of the index column in the DataFrame. can't be used together with `use_default_index`
date_name (Optional[Hashable]):
Name of the date column in the DataFrame.
date_args (Optional[Dict]):
pandas.to_datetime args used for conversion of the date column.
Expand All @@ -102,26 +106,32 @@ def __init__(
The maximum number of categories in a column in order for it to be inferred as a categorical
feature.
max_float_categories (int, default 5):
The maximum number of categories in a float column in order fo it to be inferred as a
categorical feature
The maximum number of categories in a float column in order for it to be inferred as a
categorical feature.
"""
self._data = df.copy()

# Validations
if use_index is True and index is not None:
raise DeepchecksValueError('parameter use_index cannot be True if index is given')
if label is not None:
if label.shape[0] != self._data.shape[0]:
raise DeepchecksValueError('Number of samples of label and data must be equal')
label_name = label_name or 'target'
self._data[label_name] = label

if use_default_index is True and index_name is not None:
raise DeepchecksValueError('parameter use_default_index cannot be True if index is given')

if index is not None and index not in self._data.columns:
error_message = f'index column {index} not found in dataset columns.'
if index == 'index':
error_message += ' If you attempted to use the dataframe index, set use_index to True instead.'
if index_name is not None and index_name not in self._data.columns:
error_message = f'index column {index_name} not found in dataset columns.'
if index_name == 'index':
error_message += ' If you attempted to use the dataframe index, set use_default_index to True instead.'
raise DeepchecksValueError(error_message)

if date is not None and date not in self._data.columns:
raise DeepchecksValueError(f'date column {date} not found in dataset columns')
if date_name is not None and date_name not in self._data.columns:
raise DeepchecksValueError(f'date column {date_name} not found in dataset columns')

if label is not None and label not in self._data.columns:
raise DeepchecksValueError(f'label column {label} not found in dataset columns')
if label_name is not None and label_name not in self._data.columns:
raise DeepchecksValueError(f'label column {label_name} not found in dataset columns')

if features:
difference = set(features) - set(self._data.columns)
Expand All @@ -131,16 +141,14 @@ def __init__(
'found in input dataframe.')
self._features = list(features)
else:
self._features = [x for x in self._data.columns if x not in {label, index, date}]

self._label_name = label
self._use_index = use_index
self._index_name = index
self._date_name = date
if date_args is None:
self._date_args = {}
else:
self._date_args = date_args
self._features = [x for x in self._data.columns if x not in {label_name, index_name, date_name}]

self._label_name = label_name
self._use_default_index = use_default_index
self._index_name = index_name
self._date_name = date_name
self._date_args = date_args or {}

self._max_categorical_ratio = max_categorical_ratio
self._max_categories = max_categories
self._max_float_categories = max_float_categories
Expand Down Expand Up @@ -176,21 +184,19 @@ def __init__(
def from_numpy(
cls: t.Type[TDataset],
*args: np.ndarray,
feature_names: t.Sequence[Hashable] = None,
label_name: Hashable = None,
columns: t.Sequence[Hashable] = None,
**kwargs
) -> TDataset:
"""Create Dataset instance from numpy arrays.
Args:
*args: (np.ndarray):
expecting it to contain two numpy arrays (or at least one),
first with features, second with labels.
feature_names (Sequence[Hashable], default None):
names for the feature columns. If not provided next names will
be assigned to the feature columns: X1-Xn (where n - number of features)
Numpy array of data columns, and second optional numpy array of labels.
columns (Sequence[Hashable], default None):
names for the columns. If none provided, the names that will be automatically
assigned to the columns will be: 1 - n (where n - number of columns)
label_name (Hashable, default None):
labels column name. If not provided next name will be used - 'target'
labels column name. If none is provided, the name 'target' will be used.
**kwargs:
additional arguments that will be passed to the main Dataset constructor.
Expand All @@ -200,10 +206,9 @@ def from_numpy(
Raises:
DeepchecksValueError:
if receives zero or more than two numpy arrays;
if features (args[0]) is not two dimensional numpy array;
if columns (args[0]) is not two dimensional numpy array;
if labels (args[1]) is not one dimensional numpy array;
if features array or labels array is empty;
if features and labels arrays are not of the same size;
Examples
--------
Expand All @@ -226,71 +231,55 @@ def from_numpy(
>>> dataset = Dataset.from_numpy(
... features, labels,
... feature_names=['sensor-1', 'sensor-2', 'sensor-3',],
... columns=['sensor-1', 'sensor-2', 'sensor-3',],
... label_name='labels'
... )
"""
if len(args) == 0 or len(args) > 2:
raise DeepchecksValueError(
"'from_numpy' constructor expecting to receive two numpy arrays (or at least one)."
"First array must contains the features and second the labels."
"First array must contains the columns and second the labels."
)

features_array = args[0]
features_error_message = (
"'from_numpy' constructor expecting features (args[0]) "
columns_array = args[0]
columns_error_message = (
"'from_numpy' constructor expecting columns (args[0]) "
"to be not empty two dimensional array."
)

if len(features_array.shape) != 2:
raise DeepchecksValueError(features_error_message)
if len(columns_array.shape) != 2:
raise DeepchecksValueError(columns_error_message)

if features_array.shape[0] == 0 or features_array.shape[1] == 0:
raise DeepchecksValueError(features_error_message)
if columns_array.shape[0] == 0 or columns_array.shape[1] == 0:
raise DeepchecksValueError(columns_error_message)

if feature_names is not None and len(feature_names) != features_array.shape[1]:
if columns is not None and len(columns) != columns_array.shape[1]:
raise DeepchecksValueError(
f'{features_array.shape[1]} features were provided '
f'but only {len(feature_names)} name(s) for them`s.'
f'{columns_array.shape[1]} columns were provided '
f'but only {len(columns)} name(s) for them`s.'
)

elif feature_names is None:
feature_names = [f'X{index}'for index in range(1, features_array.shape[1] + 1)]
elif columns is None:
columns = [str(index) for index in range(1, columns_array.shape[1] + 1)]

if len(args) == 1:
return cls(
df=pd.DataFrame(data=features_array, columns=feature_names),
features=feature_names, # type: ignore TODO
**kwargs
)

labels_array = None
else:
labels_array = args[1]
label_name = label_name or 'target'
columns = list(feature_names) + [label_name]

if len(labels_array.shape) != 1 or labels_array.shape[0] == 0:
raise DeepchecksValueError(
"'from_numpy' constructor expecting labels (args[1]) "
"to be not empty one dimensional array."
)

if labels_array.shape[0] != features_array.shape[0]:
raise DeepchecksValueError(
"'from_numpy' constructor expecting that features and "
"labels arrays will be of the same size"
)

labels_array = labels_array.reshape(len(labels_array), 1)
data = np.hstack((features_array, labels_array))
labels_array = pd.Series(labels_array)

return cls(
df=pd.DataFrame(data=data, columns=columns),
features=feature_names,
label=label_name,
**kwargs
)
return cls(
df=pd.DataFrame(data=columns_array, columns=columns),
label=labels_array,
**kwargs
)

@classmethod
def from_dict(
Expand Down Expand Up @@ -333,15 +322,15 @@ def copy(self: TDataset, new_data) -> TDataset:
# Filter out if columns were dropped
features = list(set(self._features).intersection(new_data.columns))
cat_features = list(set(self.cat_features).intersection(new_data.columns))
label = self._label_name if self._label_name in new_data.columns else None
label_name = self._label_name if self._label_name in new_data.columns else None
index = self._index_name if self._index_name in new_data.columns else None
date = self._date_name if self._date_name in new_data.columns else None

cls = type(self)

return cls(new_data, features=features, cat_features=cat_features, label=label, use_index=self._use_index,
index=index, date=date, convert_date_=False, max_categorical_ratio=self._max_categorical_ratio,
max_categories=self._max_categories)
return cls(new_data, features=features, cat_features=cat_features, label_name=label_name,
use_default_index=self._use_default_index, index_name=index, date_name=date, convert_date_=False,
max_categorical_ratio=self._max_categorical_ratio, max_categories=self._max_categories)

@property
def n_samples(self) -> int:
Expand Down Expand Up @@ -420,7 +409,7 @@ def index_col(self) -> t.Optional[pd.Series]:
Returns:
If date column exists, returns a pandas Series of the index column.
"""
if self._use_index is True:
if self._use_default_index is True:
return pd.Series(self.data.index)
elif self._index_name is not None:
return self.data[self._index_name]
Expand Down
4 changes: 2 additions & 2 deletions extensive_testing/run_extensive_tests.py
View file
Original file line number Diff line number Diff line change
Expand Up @@ -53,9 +53,9 @@ def has_errors(d):
train_df = pd.read_csv(f's3://{bucket_name}/{dataset}/train.csv')
test_df = pd.read_csv(f's3://{bucket_name}/{dataset}/val.csv')
metadata = json.loads(s3.Object(bucket_name, f'{dataset}/metadata.json').get()['Body'].read().decode('utf-8'))
train_ds = Dataset(train_df, label=metadata['label_name'], features=metadata['features'],
train_ds = Dataset(train_df, label_name=metadata['label_name'], features=metadata['features'],
cat_features=metadata['cat_features'])
test_ds = Dataset(test_df, label=metadata['label_name'], features=metadata['features'],
test_ds = Dataset(test_df, label_name=metadata['label_name'], features=metadata['features'],
cat_features=metadata['cat_features'])
models = list(filter(lambda x: x.key.endswith('joblib'), list(S3_BUCKET.objects.filter(Prefix=dataset).all())))
for model_obj in models:
Expand Down
2 changes: 1 addition & 1 deletion makefile
View file
Original file line number Diff line number Diff line change
Expand Up @@ -292,7 +292,7 @@ download:
$(PIP) install $(PROJECT)

jupyter: $(JUPYTER)
$(BIN)/jupyter-notebook $(args) --notebook-dir=$(NOTEBOOK_CHECKS)
$(BIN)/jupyter-notebook $(args) --notebook-dir=./notebooks

$(JUPYTER):
$(PIP) install jupyter
10 changes: 5 additions & 5 deletions notebooks/checks/distribution/train_test_drift.ipynb
View file

Large diffs are not rendered by default.

0 comments on commit 41053da

Please sign in to comment.