Ben_Leathers_Lightning_Talk_PyData_07_20_2016

Cupid Charities funds research for Neurofibromatosis (NF) by putting on fundraising events. Their main event, Cupid's Undie Run (www.CupidsUndieRun.com), is a 1-ish mile fun run in your underwear that starts and ends at a bar. These events take place in 36 cities around the US on the Saturday after Valentine's Day. Runners raise money to earn incentives like t-shirts, underwear, hats, etc. Since winter conditions can cause massive delays in shipping we are forced to ship the incentives to the runs almost a month in advance. The problem, however, is that the majority of the fundraising happens after the incentives have shipped, so predicting how many of each incentive to ship to each location is quite a challenge. It's important that every earned incentive is delivered, but we want to avoid purchasing too much extra.
I used historical data and machine learning to help me improve this prediction.

Transcript

1. Predicting the distribution of Undies An undie-rated problem www.cupidsundierun.com by: Ben Leathers PyData: 07-20-2016
2. Overview • Cupid’s Undie Run • the Problem • our solution • Next Steps
3. What is Cupid’s Undie Run?
6. Fundraising Incentives $100 $1000 $500 $250 $ $ $
7. the problem? How many undies Do we ship to each location?
8. Gender and size distributions
9. Shipping in Winter Weather
10. Fundraising Procrastination
11. 36 cities around the U.S. … On the same day.
12. Finding Predictors •  Total Raised •  Number of fundraisers •  Number of runners at previous incentive level •  Number of Men •  Number women •  Number of NF affected •  Number of teams •  Etc. … none of them were good predictors
13. The solution? My first foray into Machine Learning! In particular Random forests
14. WhAt Are Random Forests? • Ensemble method • Uses Bootstrapping • Random subset of Feature Space • Builds a forest of Decision trees
15. TOOLS import pandas as pd import numpy as np import csv import matplotlib.pyplot as plt from sklearn.ensemble import RandomForestRegressor from sklearn import cross_validation from sklearn.learning_curve import learning_curve from sklearn.metrics import mean_squared_error from sklearn.metrics import make_scorer
16. Mean SQUARED ERROR
17. NEXT steps • Principal component analysis • Lasso • Other algorithms • Address changes in incentive structure • Suggestions?
18. Thank You! www.cupidsundierun.com