solankiharsh Goto Github PK

amazon-sagemaker-architecting-for-ml

Materials for a 2-day instructor led course on applying machine learning

automating-video-editing

Automated video editing to avoid seeing my relatives in a wedding video Are you lazy and self-centered like me to hate to go through hours of video footage just to have a few find yourself in it (or a person of interest)? Perhaps you're preparing a video for a girlfriend's birthday celebration, wanting to find happy memories of you both only. Maybe you are scouring security video footage, looking to see just yourself in case you wish to see what you might have forgotten or something. Or it may be that you want to produce a highlight reel of your favorite footballer's game.

awesome-aws-certifications

☁️ A curated list of AWS certification materials: videos courses, practice exams, prep tips, etc.

awesome-generative-ai-guide

A one stop repository for generative AI research updates, interview resources, notebooks and much more!

awesome-interview-questions

:octocat: A curated awesome list of lists of interview questions. Feel free to contribute! :mortar_board:

aws_general_snippets

Some useful AWS libraries that I created or copied from the internet for day to day use.

basics

📚 Learn ML with clean code, simplified math and illustrative visuals.

capstone_sentiment_analysis

cicd-mlops

clustering-python

Different clustering approaches applied on different problemsets

coding-interview-prepration

coding-interview-university

A complete computer science study plan to become a software engineer.

complete-python-3-bootcamp

Course Files for Complete Python 3 Bootcamp Course on Udemy

cowin_vaccine_availability

Python script to check the available slots for Covid-19 Vaccination Centers from CoWIN API in India. You can get information filtered by your PIN and you can get slots for today or upcoming 7 days.

daily-coding-problem

Solutions to problems sent by dailycodingproblem.com

deep-learning-coursera

Deep Learning Specialization by Andrew Ng, deeplearning.ai.

dont_overfit

Avoid overfitting with a tiny sliver for training data Inspired by the Kaggle Don’t Overfit Challenge: Tiny Training Trial. The challenge; build the best performing model you can with a <5% training vs >95% test split with TF-IDF encodings on an Amazon multi-classification problem. With so many data hungry algorithms out there that take days or more to compute, we thought it’d be refreshing to go the other way and experiment with what can be done with extremely small and noisy datasets! Iterate and experiment with training times on the order of seconds. Our split is: Train: 1244 points Approach overview •Build Ensemble that includes multiple model categories: Logistic Regression, Random Forests, XGBoost, Adaboost, and Neural Networks. •Split the training dataset into K stratified folds. For each fold and model category, train a separate model using Grid Search. •Combine all models into ensemble using Averaging. I experimented with: 1.Which model categories to include in the ensemble 2.How many stratified folds to use: 1, 5, 10, 20, 40 3.How to build the ensemble: Averaging vs. Max voting 4.Oversampling techniques such as SMOTE and ADASYN: including models trained with SMOTE data in the ensemble worked for the Public leaderboad, but not for Private 5.Feature standardization: did not seem to improve anything. Lessons Learned Ensembling is the way to go, of course. Increasing the number of stratified folds improved performance. Improvements in training data accuracy (on validation set) did not necessarrily translate to better accuracies in the Public dataset. A prime example for this was the LR method that did not perform as well in the training validation accuracy compared to other methods such as NN. However, LR was an integral part of the overall Ensemble; whenever we removed it, the Public dataset accuracy ended up much worse. Ensembling using Averaging always worked better than Max voting. We kind of `overfitted' to the Public Leaderboard, i.e., our best performing model in Public was not the best in Private. Adding models trained with oversampled data, using either SMOTE or ADASYN, decreased accuracy in Private dataset. Gini impurity appeared to work better than Entropy for tree-based models.

eks-for-inference

Create a cluster docker build -t eksctl-cli . docker run -it eksctl-cli bash aws configure ./create-cluster.sh

face_detection_tf

fast-ml-demo

If you have no time for production level deployment, this could be used for very quick demo purpose

fastapi-code-generator

This code generator creates FastAPI app from an openapi file.

fastapi-mvc

Developer productivity tool for making high-quality FastAPI production-ready APIs.

fastapi-versioning

api versioning for fastapi web applications

fastapi_client

FastAPI client generator

fastapi_example

A demonstration of best practices for a large FastAPI project

fastapi_profiler

A FastAPI Middleware of joerick/pyinstrument to check your service performance.

flask_heroku_my_projects

Keeping record of my projects and learning flask and heroku along the way

fraud_detection

Experimentation on Fraud Detection Datasets

fraud_detection_graph_features

Fraud detection using Neo4j.

github-slideshow

A robot powered training repository :robot: