README.md

In urmanml/easyMLR: Easy Machine Learning - end to end ML worklflow in simple english language

easyMLR

End to End Machine Learning Framework in simple English language. It is created using a high level abstraction layer over Popular MLR package.

Benefits:

It has following benefits- 1) Faster time to development as all machine learning components are integrated into one end to end workflow 2) A standard framework for solving machine Learning problems 3) Low learning curve. Even a business person having no knowledge of Machine Learning can implement and solve a Machine Learning problem in less than two hours

Components:

It has all the components required for end to end Machine Learning workflow-

1) Data Preparation-

1. Reading the dataset

2. Missing value imputation

3. Outlier Treatment

4. Feature Engineering

5. Feature Selection

6. Removing Class imbalance

2) Deciding Learning Algorithm

1. Hyperparameter Tuning: By calling a simple function, it can tune set of relevant hyperparameters using crossvalidation

2. Benchmamrking: It can benchmark different learning algorithms to pick the best algorithm for the dataset as per a supplied metric

3. Analyze performance vs Thresholds: Generating graph of the model performance for different values of thresholds for different learning algorithms to select best learning algorithm and the threshold in case of Classification Problems

3) Model Training and Prediction

1. Model Training

2. Prediction using trained model

4) Model Evaluation

1. Performance evaluation: Generate summary of model performance as per different metrics like accuracy, precision, recall

Installing

install.packages("devtools")
library(devtools)
install_github("urmanml/easyMLR")

Sample Code

library(easymlr)

Load data

dataset<-read.csv("loan_data.csv")
colnames<-colnames(dataset)
target<-colnames[2]

Data Preprocessing

dataset<-missingValueImputation(dataset,target)
dataset<-outlierTreatment(dataset,target)
dataset<-featureEngineering(dataset)
dataset<-featureNormalization(dataset,target)

Specify problem type here-"classif","regr","cluster"

task<-createTask(dataset,target,type = "classif")
lrns<-createLearnerList(task$type)
#task<-featureSelection(task)
task<-treatClassImbalance(task)
lrns<-tuneLearners(lrns,task)

Benchmark Algorithms using Cross Validation

bmr<-benchmarkExperiment(lrns,task)
analyseThresholdVsPerformance(bmr)

Analyze output and select an Algorithm

lrn<-selectLearner(lrns,name = "gbm")

Model Training and Prediction

mod<-train(lrn,task = task)
pred<-prediction(mod,task,.29)

Performance Evaluation

easymlr::performance(pred)

urmanml/easyMLR documentation built on June 19, 2019, 5:09 a.m.