GBID

2016 Kaggle's Grupo Bimbo Inventory Demand Problem

Final Rank: 241

List of files:

1. script.R - xgboost models based on starter scripts with

a. more efficient lagged demand computations for last 5 weeks

b. overall frequencies for Agencies, Routes, Clients and Products.

c. 1 week lagged demand for 11th week data is computed after predicting 10th week demand

2. impute_v2.R & helper.R - xgboost models with

a. 5 week lagged demands at client-product-agency-route, client-product-agency and client-product levels;

b. weekly frequencies for Agencies, Routes, Clients and Products;

c. weekly 2-factor cross-tab counts for Agency, Client, Product and Routes;

d. added client_ID categories;

e. added product shortnames, quantity, weight & brand;

f. 2 sets of models were built - one with 1wk lagged demand for 10th week prediction, and one without for 11th week prediction.

3. impute_v6.R & helper_v5.R - xgboost models with

a. 8 week lagged demands client-product levels, 3 month, 4 month and 5 month moving averages (that skips;

b. weekly frequencies for Agencies, Routes, Clients and Products

d. added client_ID categories;

e. added product shortnames, quantity, weight & brand;

f. added agency town and state information

g. 2 sets of models were built - one with 1wk lagged demand for 10th week prediction, and one without for 11th week prediction.

4. impute_v2_v6.R - Average of results from impute_v2.R and impute_v6.R

5. BEST impute_v2_v6_results2.R - Average of results from script.R, impute_v2.R and impute_v6.R