EasyTest1

Strategy

This is the code and result for Easy1 problem. Following steps have been used to solve this problem :-

• Install and check the availability of packages.
• Importing very famous Crab data in Prof. Agresti Book - Categorical Analysis.
• Create Train and Test datasets.
• Estimate best hyper parameters using xgboost grid search cross validation with 5 fold.
• Estimate Test error using Poisson Log Likelihood metric using best parameters.
• Create plot of predicted , actual and base predicted y for Test dataset.

Head of Input Data

Sa is the response variable.

##   C S    W   Wt Sa
## 1 2 3 28.3 3.05  8
## 2 3 3 26.0 2.60  4
## 3 3 3 25.6 2.15  0
## 4 4 2 21.0 1.85  0
## 5 2 3 29.0 3.00  1
## 6 1 2 25.0 2.30  3

Distribution of Response

Good to fit poisson distribution.

Result of Model Best Parameters

I have taken subsample, max depth and shrinkage(step size) as hyper-parameters. We can also consider hyper-parameters like min child weight , col subsample and column subsample.

For Test-Error, I have used Poison log likehood as metric similar to xgboost crossvalidation eval metric. Root mean log squared error(RMLSE) can also be considered.

	Error
Model Error	2.43
Base Error	2.79

Predicted vs Actual vs Baseline Comparison