MACHINE-LEARNING

Algorithms

• Logistic Regression
• Linear regression
  • multiple linear regression
• LSTM
• Gradient descent
• KNN
• Naive Bayes
• Support Vector machine
• Decision tree
  • random forest
  • homogenity
  • gini index
  • information gain
  • detect heart desease
  • ads
  • disads :
    • overfitting
• Ensembling
  • boosting
  • stacking
  • random subspaces
  • Extreme randomized tree
• Bayesian network
• Association rule learning
• NLP
  • stop words
  • name entity recognition
  • parts of speech
  • relationships
  • tf-idf
• Anomaly detection
• association rule learning
• Semi supervised learning algo
  • Deep belief network
• K Means cluster
• XGBoost classifier
• K Fold crossvalidation

Project

• Classify malware family using K Means cluster
• Prediction analysis with different algorithms
• Anomaly detection with Isolation Forest algorithm
• Analyze time series algorithms
• Recommendation algorithms
  • Correlation based PearR
  • KNN
  • Cosine similarity
• Ensembling using in project

ML production pipeline

• Data acquisition
  • crawl
  • sources : text-based documentation, multimedia, video, audio
• ML model
  • seletec model
  • train
• fine tune model

• Batch learning

• Online learning
• Save trained model (joblib)
• Monitor system
• Update dataset and retrain model regularly
• Automate
  • collect data regularly
  • script to train model & fune tuning hyperparameters automatically, run every (week)
  • Script to deploy model to prod
  • script to evaluate model input data quality
  • Backup model

Large scale ML pipeline

Scale & ML distributed

Cloud native ML

• Google Cloud AI platform

Recommendation system

Tech stack

• Java

• Python

• Hardware : 64 bit 2+ GHzCPU, 1Gb

• Tensorflow, Keras, Scikit learn

• Colaboratory, Deepnote, Binder

• https://nbviewer.jupyter.org/github/ageron/handson-ml2/blob/master/index.ipynb

• Docker

• GPU driver

• If use GPU :

  • conda env create -f environment.yml
  • conda activate tf2
  • python -m ipykernel install --user --name=python3
  • jupyter notebook

• Weka

Machine learning

Algorithms

performance tuning

ML pipeline

ML deployment pipeline

Deep learning

Deep learning

ML | DL for mobile devices

• coreML
• tensorflow Lite
• ML kit
• Huawei AI mobile computing
• real time object classification
  • Core ML
  • CNN
  • 1,000 imagenet
  • dynamic model deployment
  • on device training
  • benchmark DL model on iOS devices

OpenCV

• Hand gesture recognition using kinect depth sensor

GPU programming with CUDA

• Parallelization and Amdahl's Law

• Stack :

  • Profiler: cProfile
  • python -m cProfile -s cumtime mandelbrot1.py > mandelbrot_profile.txt

• Set up : 64bit Intel/AMD based PC - Ubuntu LTS, 4GB RAM, NVIDIA Geforce GTX 1050 GPU +

  • Intel's Math Kernel Library (MKL)
  • macOS, Red Hat/Fedora, OpenSUSE, and CENTOS) should consult the official NVIDIA CUDA documentation (https://docs.nvidia.com/cuda/)
  • Cloud : Azure, AWS (read driver, compiler, CUDA toolkit )
  • Install : NVIDIA GPU driver
  • Set up C/C++ env
  • Install NVIDIA CUDA toolkit
  • Set up Python for GPU

• PyCUDA : memory capacity, core, transfer data between host -> device

• Scan CUDA kernel

• Streaming Multiprocessors (SMs)

• http://docs.nvidia.com/cuda/cuda-c-programming-guide/index.html#compute-capabilities

Test GPU on Cloud AWS

• AWS GPU instance

• https://towardsdatascience.com/how-to-set-up-deep-learning-machine-on-aws-gpu-instance-3bb18b0a2579

Resource

• https://www.datasetlist.com/
• https://mybinder.org/
• https://paperswithcode.com/
• https://modeldepot.io/
• https://www.visualdata.io/
• https://github.com/huggingface/knockknock

Archived repo: https://github.com/hiejulia/Machine-Learning---Deep-Learning---AI

Resource

-http://archive.ics.uci.edu/ml/index.php

• https://www.kaggle.com/datasets
• http://dataportals.org/
• https://opendatamonitor.eu/frontend/web/index.php?r=dashboard%2Findex
• https://www.quandl.com/
• https://en.wikipedia.org/wiki/List_of_datasets_for_machine-learning_research
• https://www.quora.com/Where-can-I-find-large-datasets-open-to-the-public
• https://www.reddit.com/r/datasets/
• https://registry.opendata.aws/

Case study

• Hotstar
• Netflix