def main():
scrape.find_new_games()
scrape.match_details()
teams = process.process_totals()
matches = scrape.upcoming_matches()
mlp = pickle.load(open('model/mlp_model.pkl', 'rb'))
X, y = db.get_training_csv()
scaler = StandardScaler()
X = scaler.fit_transform(X)
for m in matches:
try:
a_ts, b_ts = trueskill.rate_1vs1(teams[m[0]]['stats']['ts'],
teams[m[1]]['stats']['ts'])
g = {
'a_score': teams[m[0]]['stats']['score'],
'b_score': teams[m[1]]['stats']['score'],
'a_elo': teams[m[0]]['stats']['elo'],
'b_elo': teams[m[1]]['stats']['elo'],
'a_games': teams[m[0]]['stats']['games'],
'b_games': teams[m[1]]['stats']['games'],
'a_win': teams[m[0]]['stats']['wins'],
'b_win': teams[m[1]]['stats']['wins'],
'a_vs_record': teams[m[0]]['teams'][m[1]],
'b_vs_record': teams[m[1]]['teams'][m[0]],
'a_momentum': teams[m[0]]['stats']['momentum'],
'b_momentum': teams[m[1]]['stats']['momentum'],
'a_score_against': teams[m[0]]['stats']['score_against'],
'b_score_against': teams[m[1]]['stats']['score_against'],
'a_ts': process.win_probability(a_ts, b_ts),
'b_ts': process.win_probability(b_ts, a_ts),
}
vs_games = g['a_vs_record'] + g['b_vs_record']
match = [
process.stat_avg_diff(g['a_win'], g['a_games'], g['b_win'],
g['b_games']),
round(g['a_elo'] - g['b_elo'], 2),
process.stat_avg_diff(g['a_score'], g['a_games'], g['b_score'],
g['b_games']),
g['a_momentum'] - g['b_momentum'],
process.stat_avg_diff(g['a_vs_record'], vs_games,
g['b_vs_record'], vs_games),
process.stat_avg_diff(g['a_score_against'], g['a_games'],
g['b_score_against'], g['b_games']),
g['a_ts'],
]
match = scaler.transform([match])
outcome = mlp.predict_proba(match)
a_pred = round(outcome[0][1], 2) * 100
b_pred = round(outcome[0][0], 2) * 100
print(m[0], '(', a_pred, ')', ' vs ', m[1], '(', b_pred, ')')
except Exception as e:
print("### Error ### ", e)
def train_mlp():
X,y = db.get_training_csv()
scaler = StandardScaler()
X = scaler.fit_transform(X)
mlp = MLPClassifier(hidden_layer_sizes=(29,67), activation="relu", solver='adam',random_state=0,early_stopping=True,max_iter=500)
scores = cross_val_score(mlp, X, y, cv=5)
print(scores)
print(scores.mean())
mlp.fit(X,y)
pickle_file = open('model/mlp_model.pkl', 'wb')
pickle.dump(mlp,pickle_file)
pickle_file.close()
def search_mlp():
X,y = db.get_training_csv()s
scaler = StandardScaler()
X = scaler.fit_transform(X)
params = {
'hidden_layer_sizes': [x for x in itertools.product(range(2,50),repeat=2)],
'activation': ['tanh','relu'],
}
mlp = MLPClassifier(early_stopping=True,random_state=0,max_iter=100,solver='adam')
CV_rfc = GridSearchCV(estimator=mlp,param_grid=params, cv=5, n_jobs=-1, verbose=1)
CV_rfc.fit(X, y)
print(round(CV_rfc.best_score_,4)*100)
print(CV_rfc.best_params_)