benjamrio / horses-races-prediction Goto Github PK

Scenario

python ./bin/augment_race_music.py --input "tmp/2018-races/2018-12-*" --output tmp/2018-races-augmented/test

The script is processing 1355 json files

Actual

A file is created with the name tmp/2018-races-augmented/test containing all the historical data of the race

Expected

• tmp/2018-races-augmented/test should be used as a folder
• 1355 files should be saved in the output folder
• the output names should be the same as input names
• the output should be json files with an additional key: partants[index].augmentedMusique[indexMusique]
• Each musique can be written in json object like {position: 1, date:'2018-12-20 19:15:00', price: 6000 }

As of now, for each horse, we find all occurrences of his appearance in races, and compute the music in the order of the DataFrame given by the 'groupby(horse.genyId)'. If JSONs files are not chronologically ordered, the augmented music will also not be in chronological order.

For future use, it would be easier and faster to have the augmented music at a certain race directly inside the JSON race file.

Some races have missing horse genyId values, representing less than 1% of the total data. These data cannot be used for augmenting the music. It would be interesting to see if we could retrieve the horse identity with other attributes (canal turf id, name, ...).

As of now, a horse result position appears as it is in augmented music. It doesn't take into account how many horses starting the race, therefore being biased depending on the size of the race.

Also, the filling for disqualified or non-arriving horses seems a bit odd (as of now, we attribute 10th place for such a performance).