explainX: Explainable AI Framework for Data Scientists

We are looking for co-authors to take this project forward. Reach out @ ms8909@nyu.edu

ExplainX is a model explainability/interpretability framework for data scientists and business users.

Use explainX to understand overall model behavior, explain the "why" behind model predictions, remove biases and create convincing explanations for your business stakeholders.

Why we need model explainability & interpretibility?

Essential for:

Explaining model predictions
Debugging models
Detecting biases in data
Gaining trust of business users
Successfully deploying AI solution

What questions can we answer with explainX?

Why did my model make a mistake?
Is my model biased? If yes, where?
How can I understand and trust the model's decisions?
Does my model satisfy legal & regulatory requirements?

We have deployed the app on our server so you can play around with the dashboard. Check it out:

Dashboard Demo: http://3.128.188.55:8080/

Getting Started

Installation

Python 3.5+ | Linux, Mac, Windows

pip install explainx

To download on Windows, please install Microsoft C++ Build Tools first and then install the explainX package via pip

Installation on the cloud

If you are using a notebook instance on the cloud (AWS SageMaker, Colab, Azure), please follow our step-by-step guide to install & run explainX cloud.

Usage (Example)

After successfully installing explainX, open up your Python IDE of Jupyter Notebook and simply follow the code below to use it:

Import required module.

from explainx import * 
from sklearn.ensemble import RandomForestClassifier
from sklearn.model_selection import train_test_split

Load and split your dataset into x_data and y_data

#Load Dataset: X_Data, Y_Data 
#X_Data = Pandas DataFrame
#Y_Data = Numpy Array or List

X_data,Y_data = explainx.dataset_heloc()

Split dataset into training & testing.

X_train, X_test, Y_train, Y_test = train_test_split(X_data,Y_data, test_size=0.3, random_state=0)

Train your model.

# Train a RandomForest Model
model = RandomForestClassifier()
model.fit(X_train, Y_train)

After you're done training the model, you can either access the complete explainability dashboard or access individual techniques.

Complete Explainability Dashboard

To access the entire dashboard with all the explainability techniques under one roof, follow the code down below. It is great for sharing your work with your peers and managers in an interactive and easy to understand way.

5.1. Pass your model and dataset into the explainX function:

explainx.ai(X_test, Y_test, model, model_name="randomforest")

5.2. Click on the dashboard link to start exploring model behavior:

App running on https://127.0.0.1:8080

Explainability Modules

In this latest release, we have also given the option to use explainability techniques individually. This will allow the user to choose technique that fits their personal AI use case.

6.1. Pass your model, X_Data and Y_Data into the explainx_modules function.

explainx_modules.ai(model, X_test, Y_test)

As an upgrade, we have eliminated the need to pass in the model name as explainX is smart enough to identify the model type and problem type i.e. classification or regression, by itself.

You can access multiple modules:

Module 1: Dataframe with Predictions

explainx_modules.dataframe_graphing()

Module 2: Model Metrics

explainx_modules.metrics()

Module 3: Global Level SHAP Values

explainx_modules.shap_df()

Module 4: What-If Scenario Analysis (Local Level Explanations)

explainx_modules.what_if_analysis()

Module 5: Partial Dependence Plot & Summary Plot

explainx_modules.feature_interactions()

Module 6: Model Performance Comparison (Cohort Analysis)

explainx_modules.cohort_analysis()

To access the modules within your jupyter notebook as IFrames, just pass the mode='inline' argument.

Cloud Installation

If you are running explainX on the cloud e.g., AWS Sagemaker? https://0.0.0.0:8080 will not work.

After installation is complete, just open your terminal and run the following command.


lt -h "https://serverless.social" -p [port number]


lt -h "https://serverless.social" -p 8080

Walkthough Video Tutorial

Please click on the image below to load the tutorial:

(Note: Please manually set it to 720p or greater to have the text appear clearly)

Supported Techniques

Interpretability Technique	Status
SHAP Kernel Explainer	Live
SHAP Tree Explainer	Live
What-if Analysis	Live
Model Performance Comparison	Live
Partial Dependence Plot	Live
Surrogate Decision Tree	Coming Soon
Anchors	Coming Soon
Integrated Gradients (IG)	Coming Soon

Main Models Supported

No.	Model Name	Status
1.	Catboost	Live
2.	XGboost==1.0.2	Live
3.	Gradient Boosting Regressor	Live
4.	RandomForest Model	Live
5.	SVM	Live
6.	KNeighboursClassifier	Live
7.	Logistic Regression	Live
8.	DecisionTreeClassifier	Live
9.	All Scikit-learn Models	Live
10.	Neural Networks	Live
11.	H2O.ai AutoML	Live
12.	TensorFlow Models	Coming Soon
13.	PyTorch Models	Coming Soon

Contributing

Pull requests are welcome. In order to make changes to explainx, the ideal approach is to fork the repository then clone the fork locally.

For major changes, please open an issue first to discuss what you would like to change. Please make sure to update tests as appropriate.

Report Issues

Please help us by reporting any issues you may have while using explainX.

License

MIT

explainX/explainx