ernestynne / wids_kaggle_repo_2018

Kaggle - Predictive Analytics for Social Impact (WiDS 2018 Datathon)

Intro

This Readme file is for Datawraiths of Aotearoa team entry for the Predictive Analytics for Social Impact, hosted by Kaggle with the Women in Data Science Conference (WiDS).

The WiDS Datathon is intended to encourage women data scientists to participate in predictive analysis contests. Teams must be majority women to enter.

The challenge is to use demographic and behavioural information from survey respondents in India on their use of traditional and mobile financial services to predict the gender of respondents. Ideally, this data can then be used to help enable more Indian women to access financial services.

To learn more about the experience check out the blog from one of our contributors.

Packages

Executing this benchmark requires R along with the following packages:

• library(dplyr) # 0.7.4 data manipulation
• library(tidyr) # 0.8.0 used for data tidying in conjunction with dplyr
• library(reshape2) # 1.4.3 used for the melt function
• library(mice) # 2.46.0 multivariate imputation
• library(caret) # 6.0-78 classification and regression training
• library(ggplot2) # 2.2.1 nice plots
• library(xgboost) # 0.6.4.1 gradient boosting
• library(h2o) # 3.16.0.2 deep learning
• library(e1071) # 1.6-8 support vector machines
• library(Hmisc) # 4.1-1 labelling columns of dataset
• library(data.table) # 1.10.4-3 faster data extraction functions
• library(xlsx) # 0.5.7 reading excel data
• library(mlr) # 2.11 machine learning
• library(Boruta) # 5.2.0 feature selection
• library(VIM) # 4.7.0 visualising imputation
• library(stringr) # 1.2.0 string operation wrappers
• library(vtreat) # 1.0.2 dataframe
• library(FactoMineR) # 1.39 multivariate data mining
• library(unbalanced) # 2.0 racing for unablanced methods selection
• library(fastcluster) # 1.1.24 fast hierarchical clustering

How to

To run the benchmark,

1. Download the data
2. Modify extract_data.R to point to the training and validation data on your system.
3. Modify extract get_helper_functions.R to point to a place to save the submission
4. Train the model and make predictions by running main.R
5. Make a submission with the output file

wids_kaggle_repo_2018's Issues

removeConstantFeatures function does not exist

Called on line 31 of scripts_extract_data.R