Half way through - what's left to do Week 5

I have 6 or 7 weeks left. Which is basically 6 or 7 days + however much of my own time I can spend on it. There’s a lot left to do.

I think there are broadly 5 things for me to focus on, but I don’t think I’ll be able to achieve everything.

1. Get more data in

• Add in data from the last week
• Tokenize and stem queries so similar ones can be grouped together
• Investigate whether it makes sense to group queries with the same result sets

2. Get better results

• Account for error in the calculated relevance in a way that makes sense, or filter out documents I’m not confident about
• Make sure my testing code works properly and the metrics make sense
• Compare my implementation to PyClick’s implementation
• Change my training/test split to ensure the split is the same for each query
• Try out the DBN model

3. Devote more time to learning

• Read up more on open source libraries and similar projects
• Get better at writing idiomatic pandas code.
• Find a good stats book
• Audit the rest of the coursera Probabilistic Graphical Model’s course

4. Apply the results from the click model

• Build a tool to suggest best bets for a query (for content designers)
• Build a tool to compare two rankings (for search teams)
• Show the thing

5. Try learning to rank (if there’s time?)

• Engineer features
• Train a LTR model
• Build a prototype LTR system with the Elasticsearch LTR plugin
• Give a presentation on what I found