Code to create half of the winning voH2O submission for Analytics Vidhya's hackathon AVdatafest Xtreme ML Hack:
https://datahack.analyticsvidhya.com/contest/machine-learning-hackathon/lb
The code in thie repository was used as a 30% blend with vopani's code, which can be found here:
https://github.com/rohanrao91/AnalyticsVidhya_XtremeMLHack
My solution is in R and I used the data.table and h2o package to create my solution. Versions of either should not be important, but I used 1.10.0 and 3.11.0.3784, respectively. Run on a Macbook pro, but again, it should be hardware agnostic.
I created the spanish_holidays.csv file that is stored in the input folder using the data found at: https://www.timeanddate.com/holidays/spain/
I looked at this as a forecasting problem from that start, and thought I might approach it as Rohan did by lining up the yearly calendars and using a smoothed version of direct history at the same point in the year. But with 7 years of history and so many splits (types, category+subject), by the time I started working on a solution, I was aiming for a machine learning solution. Fortunately, this worked well, and even better, it complemented Rohan's strategy well.
I kept the feature space fairly small.
- Calendar Features
- day of week
- week of year
- day of year
- year
- Holiday Features (all binary)
- national
- local
- observance
- common local
- source: https://www.timeanddate.com/holidays/spain/
- Average Features
- average Contacts per day of week and Contact type
- average Resolution per day of week, category, and subject
- Other modeling
- used the entire training timeframe
- all H2O gbm
- separate models for Contacts and Resolution
- validation: started with Leave-One-Year-Out, finished measuring January - March 2016 (prior year from test predictions)
I started modeling with simple averages, as shown above, where the entire data set was used, and those results were applied to the entire data set. But soon I moved to a more sound version of the calculation where the impact of each record is removed from the calculation so that it is not leaked. And toward the end, I used an entire validation set to where no records of that set were used in the average calculations (similar to the actual prediction environment).
Contacts Model:
