Data is the new oil and everyone is trying to get into ML/AI/DL space. The repository here is created with an aim of exploring Machine Learning/Deep Learning algorithms on the ground level decode and understand better.
The aim is to create a master repository for a variety of algorithms implemented on real data to solve complex problem and help people who are new to start strong and understand things better.
The repository will have different sections:
This section currently covers the following topics:
- Decision Trees
- Random Forest
- Principal Component Analysis
Things to be added:
- Logistic Regression
- AUC ROC exploration for each ML Algorithm
- Clustering
Artificial Neural Network for me is a marterpiece of computing. Calculating millions of variables in such small time amazeses me. Once we understand the concept of ANN, we try out different combinations of Layers, Learning units, Learning rates which is bascially the soul the ANN we build. Apart from the hyperparameters, there are lot of other things which we must know in order to yield the best results which are Optimizers, Loss Functions and Initialization Methods for variables.
The ANN Section currently covers the following topics:
- Optimizers (Gradient Decent, Mini-Batch, Adam etc.) and difference in outputs
- Loss Functions: Theory to understand when to use what
- The Hyperparameter tuner for ANN
Things to be added:
- Working example of ANN with Hyperparameter tuning
- Different Initialization Methods
This section will cover the importand and difficult to find techniques of importing images and getting started. CNN is heavily dependent on transfer learning, exploring the algorithms such as VGG 16, Resnets and Yolo. I'm also planning to include various CNN projects such as Malaria Detection and Skin Tumor classification.
This Repository will contain
- How to import images: Unzip, create folders etc.
- Transfer Learning using VGG16, Resnets etc
- Any type of object detection using Yolo
- Malaria Detection
- Skin leasion detection
There are a lot of tasks in ML/DL which can be modularized and reused across problems. Idea behing the helper section is to create functions which can be used across algorithms and can yeiled desired results in less time.