Demo 2: Classification problem

Demo: Fairness on classification problems¶

# !pip install -e fairsense

import matplotlib.pyplot as plt
import numpy as np
import pandas as pd
from deel.fairsense.data_management.factory import from_numpy, from_pandas
from deel.fairsense.data_management.processing import one_hot_encode
from deel.fairsense.indices.confidence_intervals import with_confidence_intervals
from deel.fairsense.indices.cvm import cvm_indices
from deel.fairsense.indices.standard_metrics import disparate_impact
from deel.fairsense.indices.sobol import sobol_indices
from deel.fairsense.utils.dataclasses import IndicesInput
from deel.fairsense.utils.fairness_objective import y_true, squared_error, y_pred, classification_error
from deel.fairsense.visualization.plots import cat_plot
from deel.fairsense.visualization.text import format_with_intervals
from sklearn.tree import DecisionTreeClassifier
from sklearn.metrics import accuracy_score

1 study of the intrinsic fairness of the dataset¶

data wrangling¶

in this notebook we will work on the adult dataset.

data = pd.read_csv("data/adult.csv")
data["income"] = data["income"] == ">50K"
data = data.drop(['native-country', 'fnlwgt'], axis=1)
# data = data[data["native-country"] != "Holand-Netherlands"]
data.head()

	age	workclass	education	educational-num	marital-status	occupation	relationship	race	gender	capital-gain	hours-per-week	income
0	25	Private	11th	7	Never-married	Machine-op-inspct	Own-child	Black	Male	0	40	False
1	38	Private	HS-grad	9	Married-civ-spouse	Farming-fishing	Husband	White	Male	0	50	False
2	28	Local-gov	Assoc-acdm	12	Married-civ-spouse	Protective-serv	Husband	White	Male	0	40	True
3	44	Private	Some-college	10	Married-civ-spouse	Machine-op-inspct	Husband	Black	Male	7688	40	True
4	18	?	Some-college	10	Never-married	?	Own-child	White	Female	0	30	False

indice computation: Disparate impact¶

First we will start with computing some indices on the training data to see if the dataset is biased. The first step consist of building the IndicesInput object that stores the data. As we can set the target y_true means that we analyse the data, but this can be set to y_pred if we want to analyse predictions, or squared_error if we want to analyse the error. We can then apply preprocessing such as one_hot encoding.

indices_inputs = from_pandas(data, "income", target=y_true)
categorical_cols = list(filter(lambda col: data.dtypes[col] == "O", data.columns))
indices_inputs = one_hot_encode(indices_inputs, categorical_cols)

We then declare the indices computation functions. The results are stored in a indicesOuput object. raw value can be acessed with .values, note that 0 refers to total independence and 1 refers to total dependence.

indices_outputs = disparate_impact(indices_inputs)
indices_outputs.values

	DI
age	NaN
capital-gain	NaN
capital-loss	NaN
education	0.591701
educational-num	NaN
gender	0.640345
hours-per-week	NaN
marital-status	0.682576
occupation	0.439293
race	0.380506
relationship	0.747347
workclass	0.412769

cat_plot(indices_outputs, plot_per="index", kind="bar")
plt.show()

It is also possible to decorate any indice function with with_confidence_intervals to use bootstrapping to compute confidence intervals. We can also use the + operator to compute multiple indices simulteanously. Results with confidence intervals can be visualized either textually with format_with_intervals or 'graphically with cat_plot

di_with_ci = with_confidence_intervals(n_splits=30)(disparate_impact)
indices_outputs = di_with_ci(indices_inputs)
format_with_intervals(indices_outputs, quantile=0.05)

100%|â–ˆâ–ˆâ–ˆâ–ˆâ–ˆâ–ˆâ–ˆâ–ˆâ–ˆâ–ˆâ–ˆâ–ˆâ–ˆâ–ˆâ–ˆâ–ˆâ–ˆâ–ˆâ–ˆâ–ˆâ–ˆâ–ˆâ–ˆâ–ˆâ–ˆâ–ˆâ–ˆâ–ˆâ–ˆâ–ˆâ–ˆâ–ˆâ–ˆâ–ˆâ–ˆâ–ˆâ–ˆâ–ˆâ–ˆâ–ˆâ–ˆâ–ˆâ–ˆ| 30/30 [00:02<00:00, 11.66it/s]

	DI
age	nan [nan, nan]
capital-gain	nan [nan, nan]
capital-loss	nan [nan, nan]
education	0.60 [0.54, 0.65]
educational-num	nan [nan, nan]
gender	0.64 [0.58, 0.71]
hours-per-week	nan [nan, nan]
marital-status	0.73 [0.54, 0.84]
occupation	0.46 [0.40, 0.53]
race	0.42 [0.26, 0.56]
relationship	0.76 [0.69, 0.79]
workclass	0.35 [0.19, 0.47]

cat_plot(indices_outputs, plot_per="index", kind="bar")
plt.show()

2 train a model and analyse it's sensitivity¶

first we will split the data and then train a basic model on it.

data = data.sample(frac=1.) # shuffle data
data_train = data.iloc[:int(len(data)*0.8)]
data_test = data.iloc[int(len(data)*0.8):]

similarly we build the IndiceInput object

indices_inputs_train = IndicesInput(
    x=indices_inputs.x.iloc[:int(len(data)*0.8)], 
    y_true=indices_inputs.y_true.iloc[:int(len(data)*0.8)], 
    variable_groups=indices_inputs.variable_groups
)
indices_inputs_test = IndicesInput(
    x=indices_inputs.x.iloc[int(len(data)*0.8):], 
    y_true=indices_inputs.y_true.iloc[int(len(data)*0.8):], 
    variable_groups=indices_inputs.variable_groups
)

then we train a basic model: DecisionTree. Note that this analysis can be applied to any callable that can handle numpy array as inputs.

model = DecisionTreeClassifier()
model.fit(indices_inputs_train.x, indices_inputs_train.y_true)
train_acc = accuracy_score(indices_inputs_train.y_true, model.predict(indices_inputs_train.x))
val_acc = accuracy_score(indices_inputs_test.y_true, model.predict(indices_inputs_test.x))
print(f"train acc: {train_acc}, val acc {val_acc}")

train acc: 0.9718987536150283, val acc 0.8328385709898659

we set the model and the objective

indices_inputs_train.model = model.predict
indices_inputs_train.objective = y_pred
indices_inputs_test.model = model.predict
indices_inputs_test.objective = y_pred
indices_inputs.model = model.predict
indices_inputs.objective = y_pred

cvm_with_ci = with_confidence_intervals(n_splits=30)(cvm_indices)
di_with_ci = with_confidence_intervals(n_splits=30)(disparate_impact)
sobol_with_ci = with_confidence_intervals(n_splits=30)(sobol_indices)
indices_outputs_test = cvm_with_ci(indices_inputs_train) + di_with_ci(indices_inputs_train)#  + sobol_with_ci(indices_inputs_train)
format_with_intervals(indices_outputs_test, quantile=0.1)

100%|â–ˆâ–ˆâ–ˆâ–ˆâ–ˆâ–ˆâ–ˆâ–ˆâ–ˆâ–ˆâ–ˆâ–ˆâ–ˆâ–ˆâ–ˆâ–ˆâ–ˆâ–ˆâ–ˆâ–ˆâ–ˆâ–ˆâ–ˆâ–ˆâ–ˆâ–ˆâ–ˆâ–ˆâ–ˆâ–ˆâ–ˆâ–ˆâ–ˆâ–ˆâ–ˆâ–ˆâ–ˆâ–ˆâ–ˆâ–ˆâ–ˆâ–ˆâ–ˆ| 30/30 [01:10<00:00,  2.36s/it]
100%|â–ˆâ–ˆâ–ˆâ–ˆâ–ˆâ–ˆâ–ˆâ–ˆâ–ˆâ–ˆâ–ˆâ–ˆâ–ˆâ–ˆâ–ˆâ–ˆâ–ˆâ–ˆâ–ˆâ–ˆâ–ˆâ–ˆâ–ˆâ–ˆâ–ˆâ–ˆâ–ˆâ–ˆâ–ˆâ–ˆâ–ˆâ–ˆâ–ˆâ–ˆâ–ˆâ–ˆâ–ˆâ–ˆâ–ˆâ–ˆâ–ˆâ–ˆâ–ˆ| 30/30 [00:01<00:00, 16.35it/s]

	CVM	CVM_indep	DI
age	0.97 [0.94, 0.98]	0.24 [0.20, 0.29]	nan [nan, nan]
capital-gain	0.99 [0.95, 1.00]	0.01 [0.00, 0.02]	nan [nan, nan]
capital-loss	0.99 [0.96, 1.00]	0.00 [0.00, 0.02]	nan [nan, nan]
education	1.00 [0.97, 1.00]	0.01 [0.00, 0.04]	0.61 [0.56, 0.66]
educational-num	0.99 [0.97, 1.00]	0.05 [0.03, 0.07]	nan [nan, nan]
gender	0.99 [0.96, 1.00]	0.01 [0.00, 0.02]	0.62 [0.55, 0.69]
hours-per-week	0.97 [0.96, 1.00]	0.09 [0.05, 0.12]	nan [nan, nan]
marital-status	1.00 [0.96, 1.00]	0.01 [0.00, 0.03]	0.68 [0.57, 0.80]
occupation	0.97 [0.94, 0.99]	0.02 [0.00, 0.04]	0.47 [0.44, 0.55]
race	0.99 [0.97, 1.00]	0.00 [0.00, 0.02]	0.40 [0.30, 0.57]
relationship	0.99 [0.97, 1.00]	0.01 [0.00, 0.04]	0.75 [0.70, 0.78]
workclass	0.99 [0.96, 1.00]	0.01 [0.00, 0.03]	0.30 [0.22, 0.42]

cat_plot(indices_outputs_test, plot_per="index", kind="box", col_wrap=3)
plt.show()

indices_inputs_train._objective = classification_error
indices_inputs_test._objective = classification_error

cvm_with_ci = with_confidence_intervals(n_splits=15)(cvm_indices)
sobol_with_ci = with_confidence_intervals(n_splits=15)(sobol_indices)
indices_outputs_test_error = cvm_with_ci(indices_inputs_test)# + sobol_with_ci(indices_inputs_test)
format_with_intervals(indices_outputs_test_error, quantile=0.1)

100%|â–ˆâ–ˆâ–ˆâ–ˆâ–ˆâ–ˆâ–ˆâ–ˆâ–ˆâ–ˆâ–ˆâ–ˆâ–ˆâ–ˆâ–ˆâ–ˆâ–ˆâ–ˆâ–ˆâ–ˆâ–ˆâ–ˆâ–ˆâ–ˆâ–ˆâ–ˆâ–ˆâ–ˆâ–ˆâ–ˆâ–ˆâ–ˆâ–ˆâ–ˆâ–ˆâ–ˆâ–ˆâ–ˆâ–ˆâ–ˆâ–ˆâ–ˆâ–ˆ| 15/15 [00:11<00:00,  1.30it/s]

	CVM	CVM_indep
age	0.98 [0.93, 1.00]	0.21 [0.18, 0.27]
capital-gain	0.99 [0.95, 1.00]	0.00 [0.00, 0.01]
capital-loss	1.00 [0.96, 1.00]	0.00 [0.00, 0.02]
education	1.00 [0.96, 1.00]	0.02 [0.00, 0.04]
educational-num	1.00 [0.97, 1.00]	0.05 [0.03, 0.07]
gender	1.00 [0.96, 1.00]	0.00 [0.00, 0.02]
hours-per-week	0.97 [0.95, 1.00]	0.09 [0.02, 0.13]
marital-status	1.00 [0.97, 1.00]	0.01 [0.00, 0.03]
occupation	0.99 [0.96, 1.00]	0.01 [0.00, 0.04]
race	1.00 [0.97, 1.00]	0.00 [0.00, 0.02]
relationship	0.99 [0.97, 1.00]	0.01 [0.00, 0.03]
workclass	1.00 [0.96, 1.00]	0.01 [0.00, 0.03]

cat_plot(indices_outputs_test_error, plot_per="variable", kind="box", col_wrap=4)
plt.show()