README

A simple machine learning web server that caters for small datasets

Usage

git clone this project and run the following command:

cd devops/build
./make.sh

If you are on Windows run the following command instead:

cd devops/build
./make.ps1

This will build the ml-webapi-jersey.jar and ml-webapi-spring.jar in the devops/bin folder

To run the web server, use the following command:

java -jar devops/bin/ml-webapi-spring.jar

Now navigate to your web browser and enter the following url:

http://localhost:8900

Features

Machine Learning

• ML-SVM: SVC, SVR, one-class SVC (nu, epsilon, primal, SMO)
• ML-Trees: Isolation Forest, ID3
• ML-kNN: kNN classifier, anomaly detection (D-Distance, kth-NN-Distance, Avg-kNN-Distance, ORCA)
• ML-GLM: generalized linear model, linear model, logistic model
• ML-GaussianMixture: multivariate normal outliers, RANSAC
• ML-LM: linear model
• ML-Logistic: logistic model
• ML-LOF: local outlier factors (LOF), cluster-based local outlier factors (CBLOF), local distance-based outlier factors (LDOF), local correlation integral (LOCI)
• ML-Clustering: k-means clustering, expectation maximization clustering, one-link clustering, DBSCAN
• ML-Bayes: naive bayes classifier
• ML-DataPreparation: input discretization using k-means, input and output normalization
• ML-ANN: kohonen SOFM, MLP neural network, Adaptive Resonance Theory (ART1, Fuzzy ART, ARTMAP)
• ML-ARM: Apriori association rule mining
• ML-Reinforcement: Q-Learn, Q-Learn(lambda), SARSA, SARSA(lambda), action selection policy (SoftMax, EpsilonGreedy, Greedy, GibbsSoftMax)
• ML-FALCON: R-FALCON, TD-FALCON (Q-FALCON, S-FALCON, Q-FALCON(lambda), S-FALCON(lambda), applications (TSP, MineField, Flocking, Traffic)
• SK-Statistics: ICA (KICA or Kurtois Maximization ICA), PCA, Dynamic Time Warping (DTW)
• ML-NLP: Natural Language Processing

Search and Optimisation

• Search: nonlinear conjugate gradient search

Anomaly Detection

• one-class SVC (unsupervised)
• Gaussian Mixture (unsupervised)
• ORCA (unsupervised)
• Isolation Forest (unsupervised)
• Distance based (D-Distance, kth-NN-Distance, Avg-kNN-Distance) (unsupervised)
• LOF (unsupervised)
• CBLOF (unsupervised)
• LDOF (unsupervised)
• LOCI (unsupervised)
• RANSAC (unsupervised)
• logistic binary classifier (supervised)
• SVC (supervised)
• kNN (supervised)

Time Series Prediction

• Regression: SVR, GLM, LM, MLP
• Classification: kNN, SVC, Logistic, MLP, ARTMAP, ID3, C45
• Clustering: k-means, EM clustering, kohonen, Fuzzy ART, DBSCAN, single-link/hierarchical clustering
• Dynamic Time Warping (time series analysis: measure difference between two time series)
• Reinforcement Learning (Q-LEARN, FALCON ...)

Log Event Analytics

• ARM: Apriori rule mining
• Classification: kNN, NBC, SVC, Logistic or GLM, MLP, ARTMAP, ID3, C45
• Clustering: k-means, EM clustering, kohonen, single-link, ART1, Fuzzy ART, DBSCAN, hierarchical clustering

Spark Algorithm

• Classifiers: NBC
• Clustering: K-Means
• Anomaly Detection: Isolation Forest
• Recommender: Connection Recommendation, CWBTIAB, Content Filtering
• ARM: Apriori

Text Mining

• Topic Modeling: pLSA, Latent Semantic Allocation (LSA)
• Text Classification: NBC
• Text Clustering: K-Means

Natural Language Processing

• POS Tagging

Text Retrieval

• Filter: Porter Stemmer, Stop Word Removal, Lower Casing
• Tokenizer: English Tokenizer

Input Processing

• input transform: allow user to add, remove, numerically modify a particular attribute in the dataset before passing into algorithms
• input discretization: convert input numerical attributes to categorical attributes
• input level conversion: convert input categorical attributes to numerical attributes
• source separation: ICA which separate data sources from a mix signal (e.g., cocktail party problem)
• feature dimension reduction and restore: PCA which reduce the dimension of high-dimension data for processing by machine learning
• normalization: normalize data for machine learning preprocessing

Model Tuning

• parameter tuning: automatically tune the parameters of ML algorithm using parameter sweeping, random, or optimization search
• model selection: automatically select attributes from the data to be used in the model of the ML algorithms
• sample size selection: automatically select the training sample size

Unit Testing

• binary classifiers: supervised learning for two-class based classification
• anomaly: anomaly detection either using supervised or unsupervised approach
• classifiers: multi-label supervised classification
• regressions: regression-based forecasting
• clustering: one-link clustering, EM clustering, k-means clustering

Web Service

• ML-Jersey (Async RESTful services with jersey)
• RESTful services: project CRUD, algorithm CRUD, data CRUD, training and prediction
• ML-Tomcat (Java Servlet Container option)
• ML-Grizzly (HTTP Container option)
• ML-SpringText: rest services implemented using RESTController in spring

Web Client

• AngularJS portal for CRUD on project, ml modules and data batches
• allow user to train and run classifiers/anomaly detection/clustering/... using the web application via web browser
• flow chart to connect modules and batch for machine learning workflow construction
• experiment for comparing performance of various algorithms

Others

ML-MATSim: Agent-based Transport Simulation