Recently I saw that Facebook released Neural Prophet, a new forecasting package similar to Prophet, but built on top of Torch. Prophet is one of my favorite forecasting packages, given the ability to decompose forecasts, add in events and holidays, and take advantage of business user domain knowledge. Naturally, I was excited about hearing this new version, and on top of torch of all things! The package itself is early in development, so there’s obviously no R port yet.
The only constants in life are death, taxes, and the RStudio team continually crushing it. This time, they’ve ported Torch into R. I’m a fairly heavy tensorflow user, and coming from an R background had a steep learning curve incorporating it into my toolkit. While torch is simpler in a lot of ways (specifically, not requiring a python environment), these deep learning frameworks can be intimidating. What I hope to do here is demystify torch workflows a little bit by providing some overly simple use cases.
Building on what I was working on with my last post, where I was learning Tensorflow probability, I found that it was able to pick up the skew of simulated data pretty well, now I want to try it out on a real financial dataset.
For this, I picked the loan data from Lending Club. This is a nice dataset for this task because there’s a natural skew in the data due to defaults, where a borrower ends up paying less than the full amount they were lent.
Intro Two areas I’ve spent a lot of time in are finance and sports. In these two fields, I often hear the refrain to ‘think probabilistically’, whether that means continuing to go for it on 4th down, even if you were stuffed the last time, or getting back into a trade even though the last one blew up in your face. As Annie Duke lays out in her book Thinking In Bets, all decisions have uncertainty and you have to be able to consider where your eventual outcome fell in the distributuon of possible outcomes.
We are BACK with optimal lineups from week 11 simulations! (Better Late than Never!)
Not much commentary this week, we have the sickness roaring through our household, and just had family visiting, so didn’t have much time/energy to make changes. However, I did add some plotly wrappers to some plots, inspired by Jonathan Regenstein’s blog post on IPO exploration
If you want to review the overall code for scraping and optimizing projections, the initial post is here.
We are BACK with optimal lineups from week 10 simulations!
No changes again, mainly due to work (that, you know, I get paid for),family visiting, and working on Big Data Bowl, but I have been diving deep into modeling distributions, which I think can be a big help here.
If you want to review the overall code for scraping and optimizing projections, the initial post is here.
Setup library(data.
We are BACK with optimal lineups from week 9 simulations!
Solid week last week, I broke about even, Kenny Golladay finally put on a show, and I think I’m about to manually exclude Tyler Boyd from being placed in lineups.
No changes again, mainly due to work (that, you know, I get paid for), and working on Big Data Bowl, but I have been diving deep into modeling distributions, which I think can be a big help here.
