notes.txt

Josh’s next steps:
- Initial PCA
- Do the cointegration analysis
- Create Kaggle team - Bizzestimate! 
- AWS access
- Send Tyler book PDFs
- Contact Matt from 3DData

Tyler’s next steps:
- Create some basic plumbing code
- Study exploratory analysis notebooks
- Finish chapter 2 in Hands-On ML
- Reach out to Jan
- Take a look at Cloud9

General next steps:
- (Optional) Create a small subset of data to work with locally
- Find out where Zillow (and Trulia) gets its data per parameter
- Find a UID; parcelID may be specific only to Zillow


Potential new metrics:
- you would think that the main factor in Zillow's model is the value of neighboring homes with known value (e.g. following a recent sale.)
- boosting the importance of lot area and square footage in locations that have very homogeneous Zillow value estimates
- Market trends based on time series
- [External] More current market trends
- Q: Can deeper insight into evaluation metric be mined?


Admin thoughts:
- Suggest doing a work/play session together (remotely) each week
- At a minimum let’s walk each other thru the work we’ve done separately to learn
- Teams are capped at three… so let’s find a great third member :)
- Share code via GitHub and/or Kaggle Kernel until we get to a competitive point
- External data sources may be useful for cross-validation but rules don’t allow them in phase 2 so may be wasteful to use them for prediction in phase 1. I.e. strong results from cross-validating w/ current sales data should lead to higher ranking in the private leaderboard.
- The Kaggle CTO indicated teams who try a variety of approaches more commonly win, as opposed to drilling deeply into one approach
- Consider writing/sharing a kernel if we come up w/ something insightful