https://github.com/amogh7joshi/machine-learning-research
Research and implementations of machine learning algorithms and constructs, and deep learning models.
https://github.com/amogh7joshi/machine-learning-research
deep-learning deep-learning-algorithms machine-learning neural-network research
Last synced: 10 months ago
JSON representation
Research and implementations of machine learning algorithms and constructs, and deep learning models.
- Host: GitHub
- URL: https://github.com/amogh7joshi/machine-learning-research
- Owner: amogh7joshi
- License: mit
- Created: 2020-10-13T01:08:16.000Z (about 5 years ago)
- Default Branch: master
- Last Pushed: 2022-06-06T14:44:47.000Z (over 3 years ago)
- Last Synced: 2025-01-02T15:49:53.713Z (11 months ago)
- Topics: deep-learning, deep-learning-algorithms, machine-learning, neural-network, research
- Language: Python
- Homepage:
- Size: 175 KB
- Stars: 1
- Watchers: 1
- Forks: 1
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
- License: LICENSE
Awesome Lists containing this project
README
# Machine Learning Research
This repository is dedicated to my own machine learning research, in essence implementations of algorithms
from scratch as I attempt to learn them, construction of existing model architectures to attempt to get an understanding
of the internal functions of them, or experimentation with different algorithms and constructs to see their result.
Most of the implementations here are from scratch, using only NumPy and Matplotlib in order to use basic linear algebra operations
and visualize data, respectively, but the high-level neural network and other model construct implementations make use of
the [Tensorflow](https://github.com/tensorflow/tensorflow) and [PyTorch](https://github.com/pytorch/pytorch) libraries.
I primarily make these implementations in Python, due to not only the number of libraries but also because of the sheer ease that
is writing code in Python (at least to me). However, sometimes when testing optimization I will add constructs in C++.
While currently I have been primarily experimenting with supervised learning techniques, e.g. regression and even basic classification
using SVM/LDA, I have also tinkered with unsupervised learning, e.g. K-Means clustering (as pictured above), and PCA.
## Navigation
The directory names are quite literal, so to find different algorithms you can find the directory that contains the name
and then navigate through its subdirectories if necessary. Some directories that may need explanation (a dynamic list):
- The `nn-basic` directory contains any implementations of neural networks from when I was first working with neural networks.
For example, my initial MNIST digit classification network resides there.
## Implementations
What use is there in constructing these algorithms if they are not applied to anything? The `implementations` directory contains actual
usage of selected algorithms I've constructed on actual datasets, alongside a comparison to the [scikit-learn](https://github.com/scikit-learn/scikit-learn)
implementation of the same algorithm (on the same dataset). Some implementations to note (a growing list) include:
1. **Logistic Regression**: Implemented the Logistic Regression algorithm on the [Pima Indians Diabetes](https://www.kaggle.com/uciml/pima-indians-diabetes-database/discussion)
Dataset. On a train/test split of 0.75/0.25, the algorithm achieves an accuracy of 81.8%, compared to the scikit-learn equivalent of 82.4%.
## Experimentation
The `experimental` directory is a special directory, as anything in there is purely experimental research and often may not even function
as intended to. In these cases, I try and determine exactly what the issue is that prevents the algorithm from functioning correctly. More
information can be found in the directory of that README.
## License
All of the algorithms, constructs, and implementations in this library are licensed under the MIT License. You are free to work with them as you desire.