Generating counterfactuals for multi-class classification and regression models
This notebook will demonstrate how the DiCE library can be used for multiclass classification and regression for scikit-learn models. You can use any method (“random”, “kdtree”, “genetic”), just specific it in the method argument in the initialization step. The rest of the code is completely identical. For demonstration, we will be using the genetic algorithm for CFs.
[1]:
import dice_ml
from dice_ml import Dice
from sklearn.datasets import load_iris, fetch_california_housing
from sklearn.pipeline import Pipeline
from sklearn.preprocessing import StandardScaler, OneHotEncoder
from sklearn.model_selection import train_test_split
from sklearn.compose import ColumnTransformer
from sklearn.ensemble import RandomForestClassifier, RandomForestRegressor
import pandas as pd
[2]:
%load_ext autoreload
%autoreload 2
We will use sklearn’s internal datasets to demonstrate DiCE’s features in this notebook
Multiclass Classification
For multiclass classification, we will use sklearn’s Iris dataset. This data set consists of 3 different types of irises’ (Setosa, Versicolour, and Virginica) petal and sepal length. More information at https://scikit-learn.org/stable/datasets/toy_dataset.html#iris-plants-dataset
[3]:
df_iris = load_iris(as_frame=True).frame
df_iris.head()
[3]:
| sepal length (cm) | sepal width (cm) | petal length (cm) | petal width (cm) | target | |
|---|---|---|---|---|---|
| 0 | 5.1 | 3.5 | 1.4 | 0.2 | 0 |
| 1 | 4.9 | 3.0 | 1.4 | 0.2 | 0 |
| 2 | 4.7 | 3.2 | 1.3 | 0.2 | 0 |
| 3 | 4.6 | 3.1 | 1.5 | 0.2 | 0 |
| 4 | 5.0 | 3.6 | 1.4 | 0.2 | 0 |
[4]:
df_iris.info()
<class 'pandas.core.frame.DataFrame'>
RangeIndex: 150 entries, 0 to 149
Data columns (total 5 columns):
# Column Non-Null Count Dtype
--- ------ -------------- -----
0 sepal length (cm) 150 non-null float64
1 sepal width (cm) 150 non-null float64
2 petal length (cm) 150 non-null float64
3 petal width (cm) 150 non-null float64
4 target 150 non-null int64
dtypes: float64(4), int64(1)
memory usage: 6.0 KB
[5]:
outcome_name = "target"
continuous_features_iris = df_iris.drop(outcome_name, axis=1).columns.tolist()
target = df_iris[outcome_name]
[6]:
# Split data into train and test
datasetX = df_iris.drop(outcome_name, axis=1)
x_train, x_test, y_train, y_test = train_test_split(datasetX,
target,
test_size=0.2,
random_state=0,
stratify=target)
categorical_features = x_train.columns.difference(continuous_features_iris)
# We create the preprocessing pipelines for both numeric and categorical data.
numeric_transformer = Pipeline(steps=[
('scaler', StandardScaler())])
categorical_transformer = Pipeline(steps=[
('onehot', OneHotEncoder(handle_unknown='ignore'))])
transformations = ColumnTransformer(
transformers=[
('num', numeric_transformer, continuous_features_iris),
('cat', categorical_transformer, categorical_features)])
# Append classifier to preprocessing pipeline.
# Now we have a full prediction pipeline.
clf_iris = Pipeline(steps=[('preprocessor', transformations),
('classifier', RandomForestClassifier())])
model_iris = clf_iris.fit(x_train, y_train)
[7]:
d_iris = dice_ml.Data(dataframe=df_iris,
continuous_features=continuous_features_iris,
outcome_name=outcome_name)
# We provide the type of model as a parameter (model_type)
m_iris = dice_ml.Model(model=model_iris, backend="sklearn", model_type='classifier')
[8]:
exp_genetic_iris = Dice(d_iris, m_iris, method="genetic")
As we can see below, all the target values will lie in the desired class
[9]:
# Single input
query_instances_iris = x_test[2:3]
genetic_iris = exp_genetic_iris.generate_counterfactuals(query_instances_iris, total_CFs=7, desired_class=2)
genetic_iris.visualize_as_dataframe()
100%|█████████████████████████████████████████████████████████████████████████████████████| 1/1 [00:01<00:00, 1.16s/it]
Query instance (original outcome : 0)
| sepal length (cm) | sepal width (cm) | petal length (cm) | petal width (cm) | target | |
|---|---|---|---|---|---|
| 0 | 5.1 | 3.8 | 1.9 | 0.4 | 0 |
Diverse Counterfactual set (new outcome: 2)
| sepal length (cm) | sepal width (cm) | petal length (cm) | petal width (cm) | target | |
|---|---|---|---|---|---|
| 0 | 7.9 | 3.8 | 6.4 | 2.0 | 2 |
| 0 | 7.7 | 3.8 | 6.7 | 2.2 | 2 |
| 0 | 6.3 | 3.8 | 4.4 | 2.1 | 2 |
| 0 | 6.4 | 3.3 | 6.1 | 0.9 | 2 |
| 0 | 7.2 | 3.6 | 5.2 | 1.3 | 2 |
| 0 | 4.3 | 3.3 | 4.5 | 1.7 | 2 |
[10]:
# Multiple queries can be given as input at once
query_instances_iris = x_test[17:19]
genetic_iris = exp_genetic_iris.generate_counterfactuals(query_instances_iris, total_CFs=7, desired_class=2)
genetic_iris.visualize_as_dataframe(show_only_changes=True)
100%|█████████████████████████████████████████████████████████████████████████████████████| 2/2 [00:03<00:00, 1.56s/it]
Query instance (original outcome : 1)
| sepal length (cm) | sepal width (cm) | petal length (cm) | petal width (cm) | target | |
|---|---|---|---|---|---|
| 0 | 6.7 | 3.1 | 4.4 | 1.4 | 1 |
Diverse Counterfactual set (new outcome: 2)
| sepal length (cm) | sepal width (cm) | petal length (cm) | petal width (cm) | target | |
|---|---|---|---|---|---|
| 0 | - | - | 5.6 | 2.4 | 2.0 |
| 0 | 6.4 | - | 5.5 | 1.8 | 2.0 |
| 0 | 6.9 | - | 5.4 | 2.1 | 2.0 |
| 0 | 6.9 | - | 5.1 | 2.3 | 2.0 |
| 0 | - | 3.0 | 5.2 | 2.3 | 2.0 |
| 0 | 5.7 | - | 5.0 | 2.0 | 2.0 |
| 0 | - | 3.3 | 5.7 | 2.1 | 2.0 |
Query instance (original outcome : 1)
| sepal length (cm) | sepal width (cm) | petal length (cm) | petal width (cm) | target | |
|---|---|---|---|---|---|
| 0 | 7.0 | 3.2 | 4.7 | 1.4 | 1 |
Diverse Counterfactual set (new outcome: 2)
| sepal length (cm) | sepal width (cm) | petal length (cm) | petal width (cm) | target | |
|---|---|---|---|---|---|
| 0 | 7.2 | - | 6.0 | 1.8 | 2.0 |
| 0 | 6.5 | - | 5.1 | 2.0 | 2.0 |
| 0 | 6.9 | - | 5.7 | 2.3 | 2.0 |
| 0 | 6.8 | - | 5.9 | 2.3 | 2.0 |
| 0 | 6.4 | - | 5.3 | 2.3 | 2.0 |
| 0 | - | 2.8 | 1.5 | 2.0 | 2.0 |
| 0 | 6.9 | 3.1 | 5.4 | 2.1 | 2.0 |
Regression
For regression, we will use sklearn’s California Housing dataset. This dataset contains California house prices. More information at https://scikit-learn.org/stable/modules/generated/sklearn.datasets.fetch_california_housing.html
[11]:
housing_data = fetch_california_housing()
df_housing = pd.DataFrame(housing_data.data, columns=housing_data.feature_names)
df_housing[outcome_name] = pd.Series(housing_data.target)
df_housing.head()
[11]:
| MedInc | HouseAge | AveRooms | AveBedrms | Population | AveOccup | Latitude | Longitude | target | |
|---|---|---|---|---|---|---|---|---|---|
| 0 | 8.3252 | 41.0 | 6.984127 | 1.023810 | 322.0 | 2.555556 | 37.88 | -122.23 | 4.526 |
| 1 | 8.3014 | 21.0 | 6.238137 | 0.971880 | 2401.0 | 2.109842 | 37.86 | -122.22 | 3.585 |
| 2 | 7.2574 | 52.0 | 8.288136 | 1.073446 | 496.0 | 2.802260 | 37.85 | -122.24 | 3.521 |
| 3 | 5.6431 | 52.0 | 5.817352 | 1.073059 | 558.0 | 2.547945 | 37.85 | -122.25 | 3.413 |
| 4 | 3.8462 | 52.0 | 6.281853 | 1.081081 | 565.0 | 2.181467 | 37.85 | -122.25 | 3.422 |
[12]:
df_housing.info()
<class 'pandas.core.frame.DataFrame'>
RangeIndex: 20640 entries, 0 to 20639
Data columns (total 9 columns):
# Column Non-Null Count Dtype
--- ------ -------------- -----
0 MedInc 20640 non-null float64
1 HouseAge 20640 non-null float64
2 AveRooms 20640 non-null float64
3 AveBedrms 20640 non-null float64
4 Population 20640 non-null float64
5 AveOccup 20640 non-null float64
6 Latitude 20640 non-null float64
7 Longitude 20640 non-null float64
8 target 20640 non-null float64
dtypes: float64(9)
memory usage: 1.4 MB
[13]:
continuous_features_housing = df_housing.drop(outcome_name, axis=1).columns.tolist()
target = df_housing[outcome_name]
[14]:
# Split data into train and test
datasetX = df_housing.drop(outcome_name, axis=1)
x_train, x_test, y_train, y_test = train_test_split(datasetX,
target,
test_size=0.2,
random_state=0)
categorical_features = x_train.columns.difference(continuous_features_housing)
# We create the preprocessing pipelines for both numeric and categorical data.
numeric_transformer = Pipeline(steps=[
('scaler', StandardScaler())])
categorical_transformer = Pipeline(steps=[
('onehot', OneHotEncoder(handle_unknown='ignore'))])
transformations = ColumnTransformer(
transformers=[
('num', numeric_transformer, continuous_features_housing),
('cat', categorical_transformer, categorical_features)])
# Append classifier to preprocessing pipeline.
# Now we have a full prediction pipeline.
regr_housing = Pipeline(steps=[('preprocessor', transformations),
('regressor', RandomForestRegressor())])
model_housing = regr_housing.fit(x_train, y_train)
[15]:
d_housing = dice_ml.Data(dataframe=df_housing, continuous_features=continuous_features_housing, outcome_name=outcome_name)
# We provide the type of model as a parameter (model_type)
m_housing = dice_ml.Model(model=model_housing, backend="sklearn", model_type='regressor')
[16]:
exp_genetic_housing = Dice(d_housing, m_housing, method="genetic")
As we can see below, all the target values will lie in the desired range
[17]:
# Multiple queries can be given as input at once
query_instances_housing = x_test[2:4]
genetic_housing = exp_genetic_housing.generate_counterfactuals(query_instances_housing,
total_CFs=2,
desired_range=[3.0, 5.0])
genetic_housing.visualize_as_dataframe(show_only_changes=True)
100%|█████████████████████████████████████████████████████████████████████████████████████| 2/2 [00:02<00:00, 1.09s/it]
Query instance (original outcome : 1.5158300399780273)
| MedInc | HouseAge | AveRooms | AveBedrms | Population | AveOccup | Latitude | Longitude | target | |
|---|---|---|---|---|---|---|---|---|---|
| 0 | 4.3487 | 29.0 | 5.930712 | 1.026217 | 1554.0 | 2.910112 | 38.650002 | -121.839996 | 1.51583 |
Diverse Counterfactual set (new outcome: [3.0, 5.0])
| MedInc | HouseAge | AveRooms | AveBedrms | Population | AveOccup | Latitude | Longitude | target | |
|---|---|---|---|---|---|---|---|---|---|
| 0 | 3.6976 | - | 5.9 | 1.0 | - | 2.0 | 34.24 | -124.35 | 3.6281412000000004 |
| 0 | 6.5173 | 24.0 | 6.5 | 1.0 | - | 2.7 | 34.45 | -119.81 | 3.517150499999996 |
Query instance (original outcome : 0.8763800263404846)
| MedInc | HouseAge | AveRooms | AveBedrms | Population | AveOccup | Latitude | Longitude | target | |
|---|---|---|---|---|---|---|---|---|---|
| 0 | 2.4511 | 37.0 | 4.992958 | 1.316901 | 390.0 | 2.746479 | 33.200001 | -115.599998 | 0.87638 |
Diverse Counterfactual set (new outcome: [3.0, 5.0])
| MedInc | HouseAge | AveRooms | AveBedrms | Population | AveOccup | Latitude | Longitude | target | |
|---|---|---|---|---|---|---|---|---|---|
| 0 | 10.3682 | - | 8.1 | 1.1 | - | 2.6 | 33.61 | -117.92 | 4.94750939999999 |
| 0 | 2.9167 | 43.0 | 4.6 | 1.2 | - | 1.6 | 34.01 | -118.47 | 4.3242802999999945 |