-
Notifications
You must be signed in to change notification settings - Fork 0
/
training_data.py
182 lines (149 loc) · 5.73 KB
/
training_data.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
#training_data.py
import pandas as pd
import numpy as np
import read_player_stats
from sklearn import linear_model
from sklearn import svm
from sklearn import grid_search
from sklearn import preprocessing
from sklearn import cluster
from sklearn import cross_validation
def make_total_data(seasons=range(2004,2015),pages=[0,1], pos='rb'):
total_df = None
for season in seasons:
for page in pages:
df = read_player_stats.read_stats(season, week=0,page=page,pos=pos, scoring='PPR')
if total_df is None:
total_df = df
else:
total_df = total_df.append(df, ignore_index = True)
total_df.sort_index(by=['Name','Season'], inplace=True)
return total_df
def make_total_game_data(seasons=range(2004,2015), weeks=range(1,18),pages=[0,1], pos='rb', scoring='FD'):
total_df = None
for season in seasons:
for week in weeks:
for page in pages:
df = read_player_stats.read_stats(season, week=week,page=page,pos=pos, scoring=scoring)
if total_df is None:
total_df = df
else:
total_df = total_df.append(df, ignore_index = True)
total_df.sort_index(by=['Name','Season', 'Week'], inplace=True)
return total_df
def merge_seasons(df_season1, df_season2):
df_dropped1 = df_season1.drop(['Season', 'Team'], axis=1)
df_dropped2 = df_season2.drop(['Season', 'Team'], axis=1)
merged = pd.merge(df_dropped1,df_dropped2, on='Name', how='outer', suffixes=('_1', '_2'))
return merged
def make_training_df(total_df, seasons=range(2004,2014), ppg=False):
training_data_df = None
for season in seasons[:-1]:
df1 = total_df[total_df.Season == season]
df2 = total_df[total_df.Season == season+1]
label_df = total_df[total_df.Season == season+2]
if ppg:
labeled = pd.merge(merge_seasons(df1, df2), label_df[['Name','FFPPG']], on='Name')
else:
labeled = pd.merge(merge_seasons(df1, df2), label_df[['Name','FFP']], on='Name')
if training_data_df is None:
training_data_df = labeled
else:
training_data_df = training_data_df.append(labeled, ignore_index=True)
training_data_nadropped = training_data_df.dropna() #dont train on missing seasons
return training_data_nadropped
def data_for_projection(total_df, season=2015):
most_rec_df = merge_seasons(total_df[total_df.Season == season-2], total_df[total_df.Season == season-1])
most_rec_df.dropna(inplace=True)
return most_rec_df
def ff_projection(most_rec_df, model, normalize=False):
#project based on most recent data
X_proj = np.array(most_rec_df.drop('Name', axis=1))
if normalize:
X_proj = preprocessing.scale(X_proj)
y_proj = model.predict(X_proj)
return y_proj
def sort_and_reindex(df, col):
df = df.sort(columns=col, ascending=False)
df.index = range(1, len(df)+1)
return df
def cluster_players(df, n_clusters=3):
X_Cluster = np.array(df.drop(['Name', 'FFPPG'], axis=1))
X_Cluster = preprocessing.scale(X_Cluster)
kmean_model = cluster.KMeans(n_clusters=n_clusters)
y_clusters = kmean_model.fit_predict(X_Cluster)
df['Cluster'] = y_clusters
group_df = [df[df.Cluster == clust] for clust in range(n_clusters)]
for i, group_element in enumerate(group_df):
group_element['FFPPG'].hist(bins=50)
print('Group ' + str(i))
print('mean: ' + str(np.mean(group_element['FFPPG'])))
print('std: ' + str(np.std(group_element['FFPPG'])))
return group_df, kmean_model
def train_player_model(training_df):
X_train = np.array(training_df.drop(['Name','FFPPG'], axis=1))
y_train = np.array(training_df['FFPPG'])
parameters = {'alpha': np.logspace(-5,5,num=30)}
lin_model = grid_search.GridSearchCV(linear_model.Ridge(normalize=True), parameters, cv=10)
lin_model.fit(X_train, y_train)
return lin_model.best_estimator_
def train_svm_model(training_df, verbose=False):
#train SVM
X_train = np.array(training_df.drop(['Name','FFPPG'], axis=1))
y_train = np.array(training_df['FFPPG'])
X_scaled = preprocessing.scale(X_train)
best_score = 9999
best_params = {}
#grid search to optimize parameters of SVM
C_list = np.logspace(-5,2,num=11)
gamma_list = np.logspace(-3,1,num=11)
epsilon_list = [.1]
for c_test in C_list:
for gamma_test in gamma_list:
for epsilon_test in epsilon_list:
svm_model = svm.SVR(kernel='rbf', C=c_test, gamma=gamma_test, epsilon=epsilon_test)
scores = cross_validation.cross_val_score(svm_model, X_scaled, y_train, cv=5, scoring='mean_absolute_error')
mean_score = np.mean(scores)
if verbose:
print('params: ' + str(svm_model.get_params()))
print(mean_score)
if abs(mean_score) < abs(best_score):
best_score = mean_score
best_params = svm_model.get_params()
print('***Best Params***')
print(best_params)
print('Score:' + str(best_score))
#return a trained SVM with the best parameters we found
ret_svm = svm.SVR()
ret_svm.set_params(**best_params)
ret_svm.fit(X_scaled, y_train)
return ret_svm
def train_gbg_svm_model(X_train, y_train, verbose=False):
#train SVM
X_scaled = preprocessing.scale(X_train)
best_score = 9999
best_params = {}
#grid search to optimize parameters of SVM
C_list = np.logspace(-5,2,num=11)
gamma_list = np.logspace(-3,1,num=11)
epsilon_list = [.1]
for c_test in C_list:
for gamma_test in gamma_list:
for epsilon_test in epsilon_list:
svm_model = svm.SVR(kernel='rbf', C=c_test, gamma=gamma_test, epsilon=epsilon_test)
scores = cross_validation.cross_val_score(svm_model, X_scaled, y_train, cv=5)
mean_score = np.mean(scores)
if verbose:
print('params: ' + str(svm_model.get_params()))
print(mean_score)
if abs(mean_score) < abs(best_score):
best_score = mean_score
best_params = svm_model.get_params()
print('***Best Params***')
print(best_params)
print('Score:' + str(best_score))
#return a trained SVM with the best parameters we found
ret_svm = svm.SVR()
ret_svm.set_params(**best_params)
ret_svm.fit(X_scaled, y_train)
return ret_svm