ExpAL: Active Learning Framework for Experimentalists

The ExpAL is a simplified active learning toolbox with a focus on tackling data scarcity issue in experimental studies. It fits well with the current machine learning framework sci-kit learn and can be easily implemented by experimental scientists.

This code repository provides a versatile active learning framework for regression tasks. It includes two main classes: ExpAL and ExpALRetro, along with various sampling methods for selecting instances from a pool of unlabeled data.

Publication

Tackling Data Scarcity Challenge through Active Learning in Materials Processing with Electrospray. Advanced Intelligent Systems (2024).

Please cite our paper here

Installation

This code requires the following Python libraries to be installed:

• numpy
• sklearn (Scikit-learn)
• pandas

You can install ExpAL directly from source:

pip install git+https://github.com/FrankWanger/ExpAL.git

Usage

The ExpAL class in this code repository provides a flexible framework for active learning, including various functionalities such as data initialization, evaluation, querying, and result management. It is designed to work with regression tasks in experiments

1. Initialize the Active Learner:

   import numpy as np
   from sklearn.ensemble import GradientBoostingRegressor
   from ExpAL.core import ExpAL
   from ExpAL.methods import Random_Pick, Query_by_Committee, Greedy
   
   eval_regr = GradientBoostingRegressor()  # Create a regressor for modeling performance evaluation
   
   al = ExpAL(eval_regr=eval_regr, batch_size=1)

2. Initialize with Data and evaluate the initial performance:

   al.init(X_start, y_start, X_pool)
   al.eval()

3. Query for New Instances:

   X_query, query_idxs = al.query(Query_by_Committee)

4. Perform some lab experiments to label X_query(not through the following function!)

   y_new = Some_Labor_Intensive_Experiments(X_new)

5. Add New Instances to Training Data and remove queried ones from Pool:

   al.add_results(X_new, y_new)
   al.remove_from_pool(query_idxs)

6. Evaluate the Model again to see improvements:

   al.eval()

7. Repeat!

Methods

• init(X_start, y_start, X_pool): Initialize training data and pool.
• eval(**kwargs): Evaluate the regressor using cross-validation.
• query(query_method, **kwargs): Query the pool for new instances using a specified query method.
• add_results(X_new, y_new): Add newly labeled instances to the training set.
• remove_from_pool(query_idxs): Remove already queried instances from the pool.

The ExpALRetro Class

Additionally, this repository includes an extension of the ExpAL class called ExpALRetro. It provides the same functionality as ExpAL but with support for retrospective active learning, where labels for the potential experiment matrix are available.

Available Query Methods

The code repository includes three query methods:

1. Random_Pick: Randomly selects instances from the pool as a baseline active learning strategy.

2. Query_by_Committee: Selects instances using the Query-by-Committee (QBC) method, which relies on a committee of regressors to evaluate uncertainty.

3. Greedy: Selects instances that are the most distant from the original training set based on Euclidean distance.