search

nickojelly / grvmodel

Greyhound racing model
1 stars 0 forks source link

Develop function to generate regression result labels #6

Closed nickojelly closed 3 years ago

nickojelly commented 3 years ago

This is implemented rudimentarily in model.ipynb at the moment, however it is only currently present in the full race results, will need to develop a method of transferring this adjusted results over to the full form format

nickojelly commented 3 years ago

For example the output of the neural net will be 1x8 matrix, This should predict the relative finish position of the dogs, with 1 being 1st and 0 being last.

Can either be a straight linear difference between each, i.e.

[ 1, 0.875, 0.75, 0.625, 0.5, 0.375, 0.25, 0.125, 0]

Or some ranking depending on the difference in finish time and first time, scaled between 0 and 1.

nickojelly commented 3 years ago

Semi-implemented in model.ipynb by creating a feature of inverse of place, in a new column called "place_adjust"

     race_id      track  dist         grade  place      dog_id box  \

314539 12539914 Traralgon 525m Maiden Final 8.0 109032142 1
314535 12539914 Traralgon 525m Maiden Final 4.0 145856635 2
314537 12539914 Traralgon 525m Maiden Final 6.0 2131520028 3
314534 12539914 Traralgon 525m Maiden Final 3.0 167542442 4
314533 12539914 Traralgon 525m Maiden Final 2.0 128006011 5
314538 12539914 Traralgon 525m Maiden Final 7.0 1199430051 6
314536 12539914 Traralgon 525m Maiden Final 5.0 159734832 7
314532 12539914 Traralgon 525m Maiden Final 1.0 115367782 8 

   split_times split_margins run_time  place_adj

314539 5.06 0.20 32.31 0.125000
314535 5.06 0.20 31.53 0.250000
314537 5.06 0.20 31.70 0.166667
314534 4.90 0.04 31.31 0.333333
314533 5.04 0.18 31.09 0.500000
314538 5.04 0.18 32.13 0.142857
314536 4.95 0.09 31.58 0.200000
314532 4.86 0.00 31.08 1.000000