def predict(self, cs, df, last_test_day):
"""Predict outcomes given a Synergy Graph and last day to predict."""
test_df = df[(df['date'] > self._last_graph_day)
& (df['date'] <= last_test_day)]
test_df = combine_same_matchups(test_df)
test_df = greater_than_minute(test_df)
predict_df = self.predict_over_dates(cs, test_df)
return predict_df
def get_train_df(self, df):
"""Subset the training set on days before a given day."""
train_df = before_date_df(df, self._last_graph_day)
train_df = combine_same_matchups(train_df)
train_df = greater_than_minute(train_df)
return train_df
# com_df['correct'] = com_df['i_margin'] * com_df['prediction'] > 0
# # com_df = com_df[com_df['prediction'].notnull()]
return predict_df.reset_index(drop=True)
if __name__ == '__main__':
season = '2008'
X = read_season('matchups_reordered', season)
X = add_date(X)
# X = X[:int(len(X) * 0.3) + 1]
all_preds = pd.DataFrame()
k_folds = 1
for k in xrange(k_folds):
train_df, test_df = train_test_split(X, test_size=0.1)
train_df = combine_same_matchups(train_df)
# train_df = greater_than_minute(train_df)
# Reset index on test set to make it easier to merge later
test_df = test_df.reset_index(drop=True)
# Compute and Predict for Unweighted Graph
cs = ComputeSynergies(train_df)
cs.initialize_random_graphs(10)
# cs.simulated_annealing(200)
preds = predict_all(cs, test_df, season)
# Compute and Predict for Weighted Graph
csw = ComputeWeightedSynergies(train_df)
csw.genetic_algorithm(60, count=5)
preds_w = predict_all(csw, test_df, '2008')