def __init__(self):
self.label_translator = {
"age_1": "BC Age",
'age_2': 'RC Age',
'age_diff': 'Age DIFF',
'def_ko_1': 'RC Chin',
'def_ko_2': 'BC Chin',
'def_ko_diff': 'Chin DIFF',
'def_ko_share': 'Chin SHARE',
'def_strike_1': 'BC Striking Def',
'def_strike_2': 'RC Striking Def',
'def_strike_diff': 'Striking Def DIFF',
'def_strike_share': 'Striking Def SHARE',
'eff_grap_1': 'BC Grappling EFF',
'eff_grap_2': 'RC Grappling EFF',
'eff_grap_diff': 'Grappling EFF DIFF',
'eff_grap_share': 'Grappling EFF SHARE',
'eff_ko_2': 'RC KO EFF',
'eff_strike_1': 'BC Striking EFF',
'eff_strike_2': 'RC Striking EFF',
'eff_strike_diff': 'Striking EFF DIFF',
'eff_strike_share': 'Striking EFF SHARE',
'off_grapp_1': 'BC Grappling Off',
'off_grapp_2': 'RC Grappling Off',
'off_grapp_diff': 'Grappling Off DIFF',
'off_grapp_share': 'Grappling Off SHARE',
'off_ko_1': 'BC KO',
'off_ko_diff': 'KO DIFF',
'off_ko_share': 'KO SHARE',
'off_strike_1': 'BC Strike Off',
'off_strike_2': 'RC Strike Off',
'off_strike_diff': 'Strike Off DIFF',
'off_strike_share': 'Strike Off SHARE',
'prev_fights_1': 'BC Num Fights',
'prev_fights_2': 'RC Num Fights',
'rounds': 'Rounds'
}
print('retrieve_best_trained_models - loading best model')
mod = load('src/predictors/all/post_feat/trained_model.joblib')
print('retrieve_best_trained_models - successfully loaded best model')
print('retrieve_best_trained_models - loading best scale')
self.scale = load('src/predictors/all/post_feat/trained_scale.joblib')
print('retrieve_best_trained_models - successfully loaded best scale')
self.red_feats = retrieve_reduced_domain_features(domain='all')
self.explainer = shap.TreeExplainer(mod)
def _train_model(model, domain = 'strike'):
scale = StandardScaler()
X, Y = form_to_domain(domain = domain)
red_feats = retrieve_reduced_domain_features(domain = domain)
X = X[red_feats['features']]
scale.fit(X)
x_train = pd.DataFrame(scale.transform(X))
x_train.set_index(Y.index, inplace = True)
x_train.rename(columns = {i:j for i,j in zip(list(x_train), list(X))}, inplace = True)
dump(scale, 'src/predictors/%s/post_feat/trained_scale.joblib' % (domain))
model.fit(x_train, Y)
return model
def generate_new_win_score(data):
try:
print('generate_new_win_score - retrieving best model')
sub_model, sub_scale = retrieve_best_trained_models(domain = 'all')
print('generate_new_win_score - successfully retrieved best model')
print('generate_new_win_score - retrieving domain features')
red_feats = retrieve_reduced_domain_features(domain = 'all')
print('generate_new_win_score - successfully retrieved domain features')
print('generate_new_win_score - generating new score')
sub_score = sub_model.predict_proba(sub_scale.transform(np.array(data[red_feats['features']]).reshape(1,-1)))[0][1]
print('generate_new_win_score - successfully generated new score')
return sub_score
except Exception as e:
print('generate_new_win_score failed with %s' % (e))
raise e
def gen_win_pred_explainer(self, bout_id):
print('retrieve_best_trained_models - loading best model')
mod = load('src/predictors/all/post_feat/trained_model.joblib')
print('retrieve_best_trained_models - successfully loaded best model')
print('retrieve_best_trained_models - loading best scale')
scale = load('src/predictors/all/post_feat/trained_scale.joblib')
print('retrieve_best_trained_models - successfully loaded best scale')
explainer = shap.TreeExplainer(mod)
red_feats = retrieve_reduced_domain_features(domain='all')
bout_detail = refreshBout(bout_id)
# print('insert_new_ml_prob - forming new bout data')
bout_data = form_new_ml_odds_data(bout_id)[red_feats['features']]
new_row = pd.DataFrame(
scale.transform(
np.array(bout_data[red_feats['features']]).reshape(1,
-1))[0]).T
new_row.set_index(pd.Index([bout_id]), inplace=True)
new_row.rename(
columns={i: j
for i, j in zip(list(new_row), list(bout_data))},
inplace=True)
shap_values = explainer.shap_values(new_row)
columns, fighter_1_data, fighter_2_data = self._win_pred_graph_data(
shap_values, new_row)
fighter_1_data['name'] = bout_detail['fighterBoutXRefs'][0]['fighter'][
'fighterName']
fighter_2_data['name'] = bout_detail['fighterBoutXRefs'][1]['fighter'][
'fighterName']
return columns, fighter_1_data, fighter_2_data
from joblib import load
import pandas as pd
from training import retrieve_reduced_domain_features, form_new_ml_odds_data
from spring.api_wrappers import refreshBout
import numpy as np
domain = 'all'
print('retrieve_best_trained_models - loading best model')
mod = load('src/predictors/%s/post_feat/trained_model.joblib' % (domain))
print('retrieve_best_trained_models - successfully loaded best model')
print('retrieve_best_trained_models - loading best scale')
scale = load('src/predictors/%s/post_feat/trained_scale.joblib' % (domain))
print('retrieve_best_trained_models - successfully loaded best scale')
red_feats = retrieve_reduced_domain_features(domain=domain)
bout_id = '57939328c1214107'
bout_detail = refreshBout(bout_id)
print('insert_new_ml_prob - forming new bout data')
bout_data = form_new_ml_odds_data(bout_id)[red_feats['features']]
new_row = pd.DataFrame(
scale.transform(np.array(bout_data[red_feats['features']]).reshape(
1, -1))[0]).T
new_row.set_index(pd.Index([bout_id]), inplace=True)
new_row.rename(columns={i: j
for i, j in zip(list(new_row), list(bout_data))},
inplace=True)
explainer = shap.TreeExplainer(mod)
def __init__(self, pw):
self.pw = pw
self.weight_class_rankings = {}
self.weight_class_fighters = {}
self.weight_classes = [
'FW', 'WSW', 'WFW', 'BW', 'WBW', 'FFW', 'WFFW', 'LW', 'MW', 'WW',
'LHW', 'HW'
]
for wc in self.weight_classes:
print('initializing %s' % (wc))
self.weight_class_rankings[wc], self.weight_class_fighters[
wc] = calc_rankings_for_wc(wc)
print('initialized %s' % (wc))
self.label_translator = {
"age_1": "BC Age",
'age_2': 'RC Age',
'age_diff': 'Age DIFF',
'age_share': 'Age SHARE',
'def_grapp_1': 'BC Grappling Def',
'def_grapp_2': 'RC Grappling Def',
'def_grapp_diff': 'Grappling Def DIFF',
'def_grapp_share': 'Grappling Def SHARE',
'def_ko_2': 'BC Chin',
'def_ko_diff': 'Chin DIFF',
'def_ko_share': 'Chin SHARE',
'def_strike_1': 'BC Striking Def',
'def_strike_2': 'RC Striking Def',
'def_strike_diff': 'Striking Def DIFF',
'def_strike_share': 'Striking Def SHARE',
'def_sub_1': 'BC Submission Def',
'def_sub_2': 'RC Submission Def',
'def_sub_diff': 'Submission Def DIFF',
'def_sub_share': 'Submission Def SHARE',
'eff_grap_1': 'BC Grappling EFF',
'eff_grap_2': 'RC Grappling EFF',
'eff_grap_diff': 'Grappling EFF DIFF',
'eff_grap_share': 'Grappling EFF SHARE',
'eff_ko_1': 'BC KO EFF',
'eff_ko_2': 'RC KO EFF',
'eff_ko_diff': 'KO EFF DIFF',
'eff_ko_share': 'KO EFF SHARE',
'eff_strike_1': 'BC Striking EFF',
'eff_strike_2': 'RC Striking EFF',
'eff_strike_diff': 'Striking EFF DIFF',
'eff_strike_share': 'Striking EFF SHARE',
'eff_sub_1': 'BC Submission EFF',
'eff_sub_2': 'RC Submission EFF',
'eff_sub_diff': 'Submission EFF DIFF',
'eff_sub_share': 'Submission EFF SHARE',
'off_grapp_1': 'BC Grappling Off',
'off_grapp_2': 'RC Grappling Off',
'off_grapp_diff': 'Grappling Off DIFF',
'off_grapp_share': 'Grappling Off SHARE',
'off_ko_1': 'BC KO',
'off_ko_2': 'RC KO',
'off_ko_diff': 'KO DIFF',
'off_ko_share': 'KO SHARE',
'off_strike_1': 'BC Strike Off',
'off_strike_2': 'RC Strike Off',
'off_strike_diff': 'Strike Off DIFF',
'off_strike_share': 'Strike Off SHARE',
'off_sub_1': 'BC Submission Off',
'off_sub_2': 'RC Submission Off',
'off_sub_diff': 'Submission Off DIFF',
'off_sub_share': 'Submission Off SHARE',
'prev_fights_1': 'BC Num Fights',
'prev_fights_2': 'RC Num Fights',
'rounds': 'Rounds'
# 'def_ko_1': 'RC Chin',
# 'def_ko_2': 'BC Chin',
# 'def_ko_diff': 'Chin DIFF',
# 'def_ko_share': 'Chin SHARE',
# 'def_strike_1': 'BC Striking Def',
# 'def_strike_2': 'RC Striking Def',
# 'def_strike_diff': 'Striking Def DIFF',
# 'def_strike_share': 'Striking Def SHARE',
# 'eff_grap_1': 'BC Grappling EFF',
# 'eff_grap_2': 'RC Grappling EFF',
# 'eff_grap_diff': 'Grappling EFF DIFF',
# 'eff_grap_share': 'Grappling EFF SHARE',
# 'eff_ko_2': 'RC KO EFF',
# 'eff_strike_1': 'BC Striking EFF',
# 'eff_strike_2': 'RC Striking EFF',
# 'eff_strike_diff': 'Striking EFF DIFF',
# 'eff_strike_share': 'Striking EFF SHARE',
# 'off_grapp_1': 'BC Grappling Off',
# 'off_grapp_2': 'RC Grappling Off',
# 'off_grapp_diff': 'Grappling Off DIFF',
# 'off_grapp_share': 'Grappling Off SHARE',
# 'off_ko_1': 'BC KO',
# 'off_ko_diff': 'KO DIFF',
# 'off_ko_share': 'KO SHARE',
# 'off_strike_1': 'BC Strike Off',
# 'off_strike_2': 'RC Strike Off',
# 'off_strike_diff': 'Strike Off DIFF',
# 'off_strike_share': 'Strike Off SHARE',
}
print('retrieve_best_trained_models - loading best model')
mod = load('src/predictors/all/post_feat/trained_model.joblib')
print('retrieve_best_trained_models - successfully loaded best model')
print('retrieve_best_trained_models - loading best scale')
self.scale = load('src/predictors/all/post_feat/trained_scale.joblib')
print('retrieve_best_trained_models - successfully loaded best scale')
self.red_feats = retrieve_reduced_domain_features(domain='all')
self.explainer = shap.TreeExplainer(mod)