The auto-sklearn from pfistfl

Hi there, 👋 I am Florian!

I'm an open-source software software developer from Munich, Germany!
I currently actively develop open-source software mainly in R and Python. Before leaving academia and joining hema.to, I enjoyed researching Automated Machine Learning, Algorithmic Fairness, ML for Flow Cytometry and many other areas of machine learning. Right now, my focus is on harnessing the power of ML and flow cytometry to help better diagnose leukaemias and other disease of the blood. Here you can find my current projects and research interests!

Current Projects

R I am a member of the mlr-org core-team, we develop the mlr3 ecosystem.
I have authored/contributed to the following packages:

I furthermore worked on several other software packages:

mcboost
classiFunc (defunct)

Python

yahpo-gym
I am currently working on improving yahpo-gym, a toolbox for researchers in hyperparameter optimization.
It is built using Pytorch and ONNX and allows lightweight and lightning fast access to neural network-based surrogates.

Interests

I am interested in developing software that brings a tangible benefit to users and society.
I want to work in and learn more about the following areas:

Algorithmic fairness
Deploying ML in the real world / MLOps
LLM's in the medical space
Everything AutoML
ML for flow cytometry
Deep Learning on tabular data

Freelancing

If you are interested in working together, either on open source projects or on other projects, contact me!
I have worked in a freelance capacity on several different projects before and greatly enjoyed the challenges.
I am always interested in getting to know new people and tackling interesting problems.
You can reach me at [email protected].

Other Info

🌱 I’m currently learning more about MLOps concepts and using Kubernetes
💬 Ask me about R, ML, HPO, Few-shot/Zero-shot Optimization and Algorithmic Fairness
Reach me at [email protected]
😄 Pronouns: He/him

Auto-Sklearn & Fairness

Paging @mfeurer @eddiebergman here for discussion.

The master/main branch of this contains a very bare bones POC version of auto-sklearn for fairness.
The following script runs on my system.
It currently makes the following assumptions:

The first column in X is the sensitive attrivute.
This does not change throughout the pipeline.

import copy
import pytest
import numpy as np
import pandas as pd

def test_metrics():
    from autosklearn.metrics import acc_diff
    m = acc_diff
    assert m.needs_prot
    m = copy.copy(m)
    m._kwargs.update({'sensitive_features': np.array([0, 1, 0, 1])})
    
    assert m(np.array([0, 0, 0, 1]), np.array([0, 1, 0, 1])) == -.5
    assert m(np.array([0, 1, 0, 1]), np.array([0, 1, 0, 1])) == -1.


def test_fit():
    import autosklearn
    import autosklearn.classification
    from autosklearn.metrics import acc_diff

    from sklearn.model_selection import train_test_split
    from sklearn.datasets import fetch_openml
    from sklearn.metrics import accuracy_score, precision_score, recall_score
    from sklearn.tree import DecisionTreeClassifier
    

    # Adult dataset with protected attribute sex
    data = fetch_openml(data_id=1590, as_frame=True)
    X = pd.get_dummies(data.data)
    Y = (data.target == ">50K") * 1

    protected_variable= "sex"
    X = pd.concat([data.data[[protected_variable]], X], axis = 1)

    X_train, X_test, Y_train, Y_test = train_test_split(X, Y)

    automl = autosklearn.classification.AutoSklearnClassifier(
        time_left_for_this_task=30,
        per_run_time_limit=5,
        initial_configurations_via_metalearning=0,
        metric=acc_diff,
        include = {
            'classifier': ['fair_exponentiated_gradient'],
            'feature_preprocessor': ["dropcor"]
        }
    )
    automl.fit(
        X_train, Y_train
    )
    score = automl.score(X_test, Y_test)
    print(score)
    del automl

test_metrics()
test_fit()

pfistfl / auto-sklearn Goto Github PK

auto-sklearn's Introduction

Hi there, 👋 I am Florian!

Current Projects

Interests

Freelancing

Other Info

auto-sklearn's People

Contributors

Watchers

auto-sklearn's Issues

Auto-Sklearn & Fairness

Recommend Projects

React

Vue.js

Typescript

TensorFlow

Django

Laravel

D3

Recommend Topics

javascript

web

server

Machine learning

Visualization

Game

Recommend Org

Facebook

Microsoft

Google

Alibaba

D3

Tencent