#!/usr/bin/env python
# -*- coding: utf-8; -*-
# Copyright (c) 2020, 2022 Oracle and/or its affiliates.
# Licensed under the Universal Permissive License v 1.0 as shown at https://oss.oracle.com/licenses/upl/
import matplotlib.pyplot as plt
import numpy as np
import pandas as pd
from sklearn import preprocessing
from sklearn.feature_selection import f_regression, f_classif, SelectKBest
from ads.type_discovery.typed_feature import ContinuousTypedFeature
from ads.common.decorator.runtime_dependency import (
runtime_dependency,
OptionalDependency,
)
[docs]
class FeatureImportance:
@staticmethod
def _get_feature_ranking(sampled_df, target, target_type, score_func=None, k=None):
if score_func is None:
score_func = (
f_regression
if isinstance(target_type, ContinuousTypedFeature)
else f_classif
)
# if there are any Nans, get the most common value from each column and replace. If entire column is Nan and not categorical, replace with 0
if sampled_df.isnull().any().sum() > 0:
mode_values = sampled_df.mode().iloc[0]
for i, col in enumerate(sampled_df.columns):
if pd.isna(mode_values[i]):
if sampled_df[col].dtype.name != "category":
mode_values[i] = 0
sampled_df.fillna(mode_values, inplace=True)
# label encoding of categorical features
categorical_columns = sampled_df.select_dtypes(include=["object", "category"])
if len(categorical_columns.columns) > 0:
categorical_columns_encoded = categorical_columns.apply(
preprocessing.LabelEncoder().fit_transform
)
sampled_df[categorical_columns.columns.values] = categorical_columns_encoded
# convert datetimes to float
datetime_columns = sampled_df.select_dtypes(include=["datetime64[ns]"])
if len(datetime_columns.columns) > 0:
sampled_df[
datetime_columns.columns.values
] = datetime_columns.values.astype("float")
timestamp_columns = sampled_df.select_dtypes(include=["datetime64[ns, UTC]"])
if len(timestamp_columns.columns) > 0:
sampled_df[
timestamp_columns.columns.values
] = timestamp_columns.values.astype("datetime64[ns]").astype("float")
extracter = SelectKBest(score_func=score_func, k="all")
columns = sampled_df.columns.values
feature_names = np.delete(columns, np.argwhere(columns == target))
feature_scores = extracter.fit(
sampled_df.drop(target, axis=1), sampled_df[target]
)
feature_ranking_df = (
pd.DataFrame(
sorted(
zip(
feature_scores.scores_,
feature_names[extracter.get_support(indices=True)],
),
reverse=True,
),
columns=["scores", "features"],
)[["features", "scores"]]
.fillna(0)
.sort_values(by=["scores"], ascending=False)[:k]
)
return feature_ranking_df
def __init__(self, ds, score_func=None, n=None):
self.feature_ranking = self._get_feature_ranking(
ds.sampled_df, ds.type_of_target(), score_func=score_func, k=n
)
def __repr__(self):
return str(self.feature_ranking)
[docs]
@runtime_dependency(module="seaborn", install_from=OptionalDependency.VIZ)
def show_in_notebook(self, fig_size=(10, 10)):
"""
Shows selected features in the notebook with matplotlib.
"""
if not utils.is_notebook():
print("show_in_notebook called but not in notebook environment")
return
with seaborn.axes_style(style="whitegrid"):
plt.figure(figsize=fig_size)
seaborn.barplot(
x="scores",
y="features",
data=self.feature_ranking,
orient="h",
palette="Blues_d",
)