Ditching R for Python

class: center, middle, inverse, title-slide

# Ditching R for Python
## (in Algorithmic Trading)
### Clayton Yochum
### PyData Ann Arbor - October 10, 2017

---

# Who the hell am I?

--
- data science engineer @ Methods Consultants

--
- co-organizer of Ann Arbor R User Group

--
- co-maintainer of `rdrop2` package for using Dropbox from R

--
- diehard R fan

--
- not really a Python user

---

# The Problem

--
- finance client w/ killer data

--
- turn that data into a _trading strategy_

--
- be sure strategy is worth using

--
- (too proprietary to show much detail)

---

# The Process

--
- raw data ->

--
- clean features ->

--
- predictions from model(s) ->

--
- picks ->

--
- position changes ->

--
- evaluation

---

# What I was doing...

--
- writing a ton of custom R code

--
- so much code

--
- poorly tested code

--
- limited functionality

---

# But Then!

--
- PyData meetup, July 2017

--
- Gus Gordon from Quantopian

--
- open-source tools: `AlphaLens`, `Zipline`, `Pyfolio`

--
- was very clear they were doing a much better job of solving the same problems

---

# Incremental Change

--
- `Pyfolio` to evaluate existing backtests

--
- `Zipline` to turn _predictions_ into _evaluated results_

--
- (haven't really used `AlphaLens` yet)

---

# Half & Half

--
- still doing modeling in R

--
- predictions in Zipline would take _too damn long_

---

# Benefits

--
- using battle-tested code

--
- so many more metrics & plots

--
- nice decoupling of _prediction_ from _strategy_

---

# Challenges

--
- Python is actually pretty different from R

--
- mixed codebases can be a pain

--
- `tidyverse` >>> `pandas`

--
- why do objects change type when you subset to one element?!?

---

# What's Next

--
- `Pachyderm`!

--
    - automated workflow

--
    - easily mix languages

--
    - Docker containers everywhere

---
class: middle

# Lessons Learned

--
#### 1. use the right tool for the job

--
#### 2. learn from others

--
#### 3. 'NIH' will kill you

---
class: center, middle

# Thanks!

@claytonyochum

claytonjy@gmail.com

`claytonjy` on Github