Applied Machine Learning and Implementation

Overview

12 weeks, 2 hours / per week

20 min per episode, so six episodes per week.

This course will cover:

***** Spark MLlib

**** ML Pipeline and GraphX

*** Spark Core and Spark SQL

** Spark Streaming

* Scikit-learn for reference.

Textbooks

Advanced Analytics with Spark
Machine Learning with Spark
The Lion Way: Machine Learning plus Intelligent Optimization
Others...

week 1 Introduction

Spark ABC
Machine learning ABC
Graph Computing ABC
Demos for Spark, MLlib, and GraphX

week 2 Generalized Linear Model

Logistic regression
Linear regression
SVM
LASSO
Ridge regression
Applied demos such as Handwritten digits recognition, etc.

week 3 Recommendation

Recommendation ALS
Singular Value Decomposition
The implementation in both MLlib and Mahout
Applied demo of recommendation with PredictionIO.

week 4 Clustering

k-means
LDA
Applied demo of geo-location clustering and topic modeling

week 5 Streaming-wised Machine Learning

Lambda Architecture
Parameter Server
Several algorithms from Freeman labs
Applied demo such as the zebrafish experiment

week 6 ML Pipeline

Pipeline of Scikit-learn
Pipeline of Spark (DataFrame, ML Pipeline, etc.)
Applied demo (TBD)

week 7 Scientific Computing

Scientific computing and Notices from Matrix Computation
Matrix libs (in C/Fortran and Java)
Matrix in MLlib
Applied demo (TBD)

week 8 The Graph Computation Model

Graph computing and libs
revisit LDA, ALS
Applied demo such as community detection for food network/recommendation.

week 9 Tree Model and Boosting

Tree model
Random forest
Ensemble in Kaggle and practice
Applied demo for ensemble

week 10 Evaluation

Evaluation methods
Implementations in MLlib
Online / Offline evaluations

week 11 Optimization in Parallel

Commonly used optimization algorithms
Sequential gene of optimization algorithms
BSP model to BSP+ model to SSP
Future ways?

week 12 Rethink of practical machine learning and how to build a good system

One, two, three of practical ML
Rethink of practical machine learning
How to build a great machine learning system?
Compare with Mahout / Oryx2 / VM / ...

Survey of Advanced Analytics with Spark

Structure of directories

/src/chapterx --> The code snippets of each chapter

/src/chapterx/{java, python, scala} --> Code snippets written with Mahout, Scikit-learn, and Spark

Spark VS Scikit-learn

Algorithms

Type	Algorithm	Scikit-learn	Spark
Classification	Logistic Regression	YES	YES
Classification	Perceptron	YES
Classification	Passive Aggressive Algorithms	YES
Classification	SVM	YES	YES
Classification	Naive Bayes	YES	YES
Classification	Decision Tree	YES	YES
Classification	Ensemble methods	YES	YES
Classification	Label Propogation	YES	YES (in GraphX)
Classification	LDA and QDA	YES
Regression	Ordinary Least Square	YES	YES
Regression	Ridge Regression	YES	YES
Regression	LASSO	YES	YES
Regression	Elastic Net	YES
Regression	Multi-task LASSO	YES
Regression	Least Angle Regression	YES
Regression	LARS LASSO	YES
Regression	Orthogonal Matching Pursuit	YES
Regression	Bayesian Regression	YES
Regression	Polynomial Regression	YES
Regression	Nearest Neighbor	YES	YES
Regression	Gaussian Process	YES
Regression	Isotonic Regression	YES
Clustering	K-means	YES	YES
Clustering	Affinity Propagation	YES
Clustering	Mean shift	YES
Clustering	Spectral Clustering	YES
Clustering	Ward	YES
Clustering	Agglomerative clustering	YES
Clustering	DBSCAN	YES
Clustering	Gaussian Mixtures	YES
Dimension Reduction	PCA	YES	YES
Dimension Reduction	SVD / LSA	YES	YES
Dimension Reduction	Dictionary Learning	YES
Dimension Reduction	Factor Analysis	YES
Dimension Reduction	ICA	YES
Dimension Reduction	NMF	YES
Model Selection	Cross Validation	YES	YES
Model Selection	Grid Search	YES
Model Selection	Pipeline	YES	YES
Model Selection	Feature Union	YES	YES
Model Selection	Model Evaluation	YES	YES
Model Selection	Model Presistence	YES
Model Selection	Validation Curves	YES
Preprocessing	Standardization	YES	YES
Preprocessing	Encoding categorical features	YES	YES (dependency)
Preprocessing	Binarization	YES
Preprocessing	Normalization	YES	YES
Preprocessing	Label preprocessing	YES
Preprocessing	Imputation of missing values	YES
Preprocessing	Unsupervised data reduction	YES

mindis / amli Goto Github PK

amli's Introduction