python at yhat (august 2013)

26
Python at Yhat Dev StackUp August 2013

Upload: yhat

Post on 26-Jan-2015

108 views

Category:

Technology


1 download

DESCRIPTION

 

TRANSCRIPT

Page 1: Python at yhat (august 2013)

Python at YhatDev StackUpAugust 2013

Page 2: Python at yhat (august 2013)

Agenda

About Yhat

How we use Python

Questions

Page 3: Python at yhat (august 2013)

We need to reduce churn. Okay. I'll look into it.

Lots of conversations like this

Page 4: Python at yhat (august 2013)

I figured out that....some complex stuff about vector space that'll improve...

....and that's how we'll reduce churn.

Sounds good. Let's do that...

The "a ha" moment isn't the end.

Page 5: Python at yhat (august 2013)

Now what?

Any of you know what Gradient Boosting is?

So when can we go live with the new model?

Page 6: Python at yhat (august 2013)
Page 7: Python at yhat (august 2013)

What goes on in the Kludge?

Rewriting CodeBatch JobsPMML

Page 8: Python at yhat (august 2013)

How can we...

- eliminate implementation time

Page 9: Python at yhat (august 2013)

How can we...

- eliminate implementation time - let data scientists use their favorite tools

Page 10: Python at yhat (august 2013)

How can we...

- eliminate implementation time - let data scientists use their favorite tools

...without altering your workflow

Page 11: Python at yhat (august 2013)

How do we do this?

Page 12: Python at yhat (august 2013)

How do we do this?

Page 13: Python at yhat (august 2013)

great for analysis● Built for analysis and statistics● Everything is tabular● Active community; 4000+ packages

Page 14: Python at yhat (august 2013)

great for analysis● Built for analysis and statistics● Everything is tabular● Active community; 4000+ packages

bad for applications● Not web friendly● Everything is tabular● Slow● A list of R grievances:

○ https://github.com/tdsmith/aRrg

Page 15: Python at yhat (august 2013)

Hooking R up to Python

Page 16: Python at yhat (august 2013)

R code

Page 17: Python at yhat (august 2013)

R code > Compile to Bytecode

Page 18: Python at yhat (august 2013)

R code > Compile to Bytecode > Execute from Python

{ “data”: {

“foo”: 100, “bar”: 200

}}

Incoming data for prediction Make prediction

from Python using compiled R

Page 19: Python at yhat (august 2013)

R code > Compile to Bytecode > Execute from Python

Returned via REST API

Prediction sent back to Python webserver{

“prob”: 0.95}

Page 20: Python at yhat (august 2013)

{}

approach

Same Python server

Page 21: Python at yhat (august 2013)

{}

approach

Plug in different scientific environments

Page 22: Python at yhat (august 2013)

{ } “prob”: 0.87

approach

Predictions sent back up the chain and to the client

Page 23: Python at yhat (august 2013)

Result

● Ensures cross environment validation● Extensible to other languages

Page 24: Python at yhat (august 2013)

Want to try?yhathq.com