Merge pull request #20 from AryaXAI/image_metrics_03022025

Added Batchwise Metric Calculation ; Cache Batchwise Metric Storage; Bug Fixes; Tqdm

Author: chintanarya

README.md

@@ -31,40 +31,9 @@
 To install **`xai_evals`**, you can use `pip`. First, clone the repository or download the files to your local environment. Then, install the necessary dependencies:
 
 ```bash
-git clone https://github.com/AryaXAI/xai_evals.git
-cd xai_evals
-pip install .
+pip install xai_evals
 ```
 
-Alternatively, if you don't want to clone the repo manually, you can install the package directly from pip (after we publish it [TODO]).
-
-### Dependencies
-
-- **`dl_backtrace`**: A library for analyzing neural networks by tracing the relevance of each component from output to input.
-- **`shap==0.46.0`**: A library for computing SHAP values (SHapley Additive exPlanations) to interpret machine learning models.
-- **`lime==0.2.0.1`**: A library for generating LIME (Local Interpretable Model-Agnostic Explanations) to explain individual predictions.
-- **`xgboost==2.1.3`**: A gradient boosting library designed for efficient and scalable machine learning tasks.
-- **`scikit-learn==1.3.2`**: A versatile machine learning library for building and evaluating models, and performing preprocessing.
-- **`torch`**: A deep learning framework (PyTorch) for creating and training neural networks.
-- **`pandas==2.1.4`**: A powerful library for data manipulation and analysis, particularly useful for handling tabular data.
-- **`numpy==1.26.4`**: A fundamental library for numerical computations, providing support for large, multi-dimensional arrays and matrices.
-- **`catboost==1.2.7`**: A gradient boosting library that is particularly effective for categorical feature handling.
-- **`lightgbm==4.5.0`**: A fast, distributed, high-performance gradient boosting framework for machine learning.
-- **`tensorflow==2.14.0`**: An open-source library for machine learning and deep learning tasks, developed by Google.
-- **`captum==0.7.0`**: A model interpretability library for PyTorch, offering various attribution methods.
-- **`tf-explain`**: A library for interpreting deep learning models, particularly for TensorFlow/Keras-based models.
-- **`quantus`**: A library for evaluating model explanations using various quantitative metrics.
-
-This list ensures all required dependencies, with specific versions, are included for optimal compatibility.
-
-To install all dependencies, run:
-
-```bash
-pip install -r requirements.txt
-```
-
----
-
 ## Usage
 
 Supported Machine Learning Models for `SHAPExplainer` and `LIMEExplainer` class is as follows : 
@@ -969,7 +938,7 @@ This project is licensed under the MIT License - see the [LICENSE](LICENSE) file
 
 ### Future Plans
 
-In the future, we will continue to improve this library by:
+In the future, we will continue to improve this library.
 
 
 

requirements.txt

@@ -9,6 +9,7 @@ catboost==1.2.7
 lightgbm==4.5.0
 tensorflow==2.14.0
 captum==0.7.0
+quantus==0.5.3
 tf-explain
-quantus
 torch
+tqdm

xai_evals/explainer.py

@@ -1,4 +1,3 @@
-# shap_lime_explainer/explainer.py
 import shap
 import lime.lime_tabular
 import numpy as np
@@ -958,4 +957,4 @@ def _format_relevance(self, instance, relevance):
 
         # Sort by relevance to understand the most influential features
         relevance_df = relevance_df.sort_values(by="Attribution", key=abs, ascending=False)
-        return relevance_df
+        return relevance_df