COMPAS Example#
This example highlights the most important features that the AI model considered while predicting the likelihood of
criminal recidivism. The COMPAS Dataset is used to capture the characteristics
of criminals in Broward (Florida) between years 2013 and 2014.
This dataset can be obtained using the load_compas()
function:
>>> from xaiographs.datasets import load_compas
>>> df_dataset = load_compas()
>>> df_dataset.head(3)
id FirstName LastName Gender Age_range Ethnicity days_b_screening_arrest c_jail_in c_jail_out Days_in_jail c_charge_degree c_charge_desc is_recid is_violent_recid score_risk_recidivism score_text_risk_recidivism score_risk_violence score_text_risk_violence Low_Recid Medium_Recid High_Recid No_Recid score_text_risk_violence Low_Recid Medium_Recid High_Recid No_Recid Recid predict_two_year_recid real_two_year_recid
0 1 miguel hernandez Male Greater than 45 Other -1.0 13/8/13 6:03 14/8/13 5:41 1 F Aggravated Assault w/Firearm 0 0 1 Low 1 Low 1 0 0 1 Low 1 0 0 1 0 0 0
1 3 kevon dixon Male 25 - 45 African-American -1.0 26/1/13 3:45 5/2/13 5:36 10 F Felony Battery w/Prior Convict 1 1 3 Low 1 Low 1 0 0 0 Low 1 0 0 0 1 0 1
2 5 marcu brown Male Less than 25 African-American NaN NaN NaN 0 F Possession of Cannabis 0 0 8 High 6 Medium 0 0 1 1 Medium 0 0 1 1 0 1 0
To determine the explainability of this dataset, XAIoGraphs provides a dataset that has already been discretized and
columns with targets probabilities using
load_compas_discretized() function:
>>> from xaiographs.datasets import load_compas_discretized
>>> df_dataset, features_cols, target_cols, y_true, y_predict = load_compas_discretized()
>>> df_dataset.head(3)
id Gender Age_range Ethnicity MaritalStatus c_charge_degree is_recid is_violent_recid High_Recid Medium_Recid Low_Recid y_true y_predict
0 1 Male Greater than 45 Other Single F NO NO 0 0 1 No_Recid No_Recid
1 3 Male 25 - 45 African-American Single F YES YES 0 0 1 Recid No_Recid
2 5 Male 25 - 45 Other Separated M NO NO 0 0 1 No_Recid No_Recid
Code Example#
The following entry point (with Python virtual environment enabled) is used to demonstrate this example.
>> compas_example
Alternatively, you may run the code below to view a full implementation of all XAIoGraphs functionalities with this Dataset:
from xaiographs import Explainer
from xaiographs import Why
from xaiographs import Fairness
from xaiographs.datasets import load_compas_discretized, load_compas_why
LANG = 'en'
# LOAD DATASETS & SEMANTICS
df_compas, feature_cols, target_cols, y_true, y_predict = load_compas_discretized()
df_values_semantics, df_target_values_semantics = load_compas_why(language=LANG)
# EXPLAINER
explainer = Explainer(importance_engine='LIDE', verbose=1)
explainer.fit(df=df_compas, feature_cols=feature_cols, target_cols=target_cols)
# WHY
why = Why(language=LANG,
explainer=explainer,
why_values_semantics=df_values_semantics,
why_target_values_semantics=df_target_values_semantics,
verbose=1)
why.fit()
# FAIRNESS
f = Fairness(verbose=1)
f.fit(df=df_compas[feature_cols + [y_true] + [y_predict]],
sensitive_cols=['Ethnicity', 'Gender', 'Age_range'],
target_col=y_true,
predict_col=y_predict)
XAIoWeb COMPAS#
After running the .fit() methods of each of the classes (one, two, or all three), a sequence of JSON files are
generated in the xaioweb_files folder to visualized in XAIoWeb interface.
To launch the web (with the virtual environment enabled), run the following entry point:
>> xaioweb -d xaioweb_files -o -f
And the results seen in XAIoWeb are the following:
Global Explainability#
Local Explainability#
Fairness#
