almartin82 / projprep Goto Github PK

probably many things are going to break. write tests!

convert rate stats to counting before zscoring, per luke's point

ie razzball steamer, which has a projection for both matt duffy as a giant and a texan?!

write a data cleaning function to return the projection with more AB/IP.

it has an expansive player universe and good metadata. add it to the universal_metadata vignette and the id_map.

R CMD check yells about non-ascii characters:

* checking data for non-ASCII characters ... WARNING
  Warning: found non-ASCII strings
  'Pedro B<c3><a1>ez' in object 'id_map'
  'Manny Ba<c3><b1>uelos' in object 'id_map'
  'Aar<c3><b3>n Brooks' in object 'id_map'
  'Arqu<c3><ad>medes Caminero' in object 'id_map'

this can apparently be solved by adding encoding info.
@chrishaid thought you might find this tidbit about CMD check interesting.

once we clean up team names, this will be easy enough.

I'm getting the following error message: "Error in names(df)[2] <- "fg_note" :
'names' attribute [2] must be the same length as the vector [1]"

because it doesn't project saves, all save zscores are NaN, which causes lots of problems. fix that.

for instance, kershaw has negative ERA value. yeah.

eg, per cheatsheet, some projections are good on true talent but bad at getting AB / IP right.

allow users to treat the playing time assumptions from one projection system as source-of-truth.

what I thought was an error in the projections for #26 is actually... just how things are going. fangraphs (a reasonably complete but not outlandish projection universe) has 13 name collsions:

table(pos_df$fullname)[table(pos_df$fullname) > 1] %>% sort()

     Carlos Perez    Carlos Sanchez Christian Vazquez  Daniel Robertson 
                2                 2                 2                 2 
    Jesus Montero     Jose Gonzalez     Jose Martinez        Juan Perez 
                2                 2                 2                 2 
       Matt Duffy    Michael Miller    Michael Taylor     Tanner Murphy 
                2                 2                 2                 2 
      Tyler Moore 
                2 

I think our playerid match functions should pre-process these names, and attempt to match on name/team...

these are stubs right now (ex), but to combine data from multiple projection systems the real mlbids are needed.

scrape them

include force_numeric logic in the cleaning function, similar to cbs

@drewgriffith15 has some code that does this for 2015:
https://github.com/drewgriffith15/MLB/blob/master/xDraft.R

per zach sanders: http://www.fangraphs.com/fantasy/basebal-fantasy-value-above-replacement/ pt 3

to allow users to interactively change the settings used by projprep.

per issue #15 (last issue to close for projprep 0.3!), it's important to get positional eligibility right. because I am loathe to do anything manually when it is scriptable, let's pull these down from from the yahoo API.

this seems simple enough - player status is
select * from fantasysports.players where game_key = 357
using yahoo's query language, and that can be wrapped into a RESTful call by hitting
https://query.yahooapis.com/v1/public/yql?q=select%20*%20from%20fantasysports.games%20where%20game_key%3D%22357%22&format=json&diagnostics=true&callback=

roughly:

• figure out how to hit that endpoint using httr
• process the json
• figure out the pagination of the api (currently returns 25 players a go)
• store everything so it can be used by projprep

whoops. need to clarify that.

another paid_projection parser...

next task is to adjust for replacement-level, per position
https://github.com/almartin82/projection_prep/blob/dc450c07833c1b1e8b26b0c7c8813483fa1a20ea/R/proj_prep.R#L64

ie, if you are a yahoo league, apply the yahoo position eligibility, no matter what the projection system thinks.

needs to key off of mlbid.

use zach's formula
http://www.fangraphs.com/fantasy/value-above-replacement-part-three/

this creates serious inflation, because guys can be kept under their value -- pushes prices up.

it's definitely worth something. but what?

similar to #1, this will help us sanity-check our prices

per #31, we will likely need to use team names as part of the match criteria. so that means we need to normalize `c('Athletics', 'OAK', ...)

will help sanity check top X by zscore vs top X by consensus rankings

1/0 is confusing.