-
Notifications
You must be signed in to change notification settings - Fork 0
/
predictions.py
159 lines (137 loc) · 6.7 KB
/
predictions.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
import rumrunner
from rumrunner import Dataset
import random, time, json, os, math, sys
from evaluator import slate_probabilities_from_rum
from cleaner import getProbsFromDsiAndDssum
from MNL import *
from pcmpMNL import train_MNLpcmc, pred_from_MNLpcmc
from utils import *
import numpy as np
def RMSE(pred_Dsi, real_Dsi, ssize):
error = 0
assert pred_Dsi.keys() == real_Dsi.keys()
for key in real_Dsi.keys():
error += (real_Dsi[key] - pred_Dsi[key]) * (real_Dsi[key] - pred_Dsi[key])
return math.sqrt(error / ssize)
def get_probs_from_list(l):
Dsi = {}
Ds_sum = {}
for v in l:
slate = tuple(v[0])
winner = v[1]
Dsi[(slate, winner)] = Dsi.get((slate, winner), 0) + 1
Ds_sum[slate] = Ds_sum.get(slate, 0) + 1
return getProbsFromDsiAndDssum(Ds_sum, Dsi, 0)[0]
def get_sums_from_list(l):
Dsi = {}
for v in l:
slate = tuple(v[0])
winner = v[1]
Dsi[(slate, winner)] = Dsi.get((slate, winner), 0) + 1
return Dsi
def get_statistics(dir):
error_list = []
for f in os.listdir(dir):
if f.split('_')[-1] == 'result':
with open(dir+'/'+f, 'r') as ff:
error = json.load(ff)['error']
error_list += [error]
errornp = np.array(error_list)
return np.mean(errornp), np.std(errornp)
def predict_via_training_data(train_probs, test_slates, slate_size):
preds = {}
for slate in test_slates:
for i in slate:
preds[(slate, i)] = train_probs.get((slate, i), 1/slate_size) #uniform probability for out-of-dataset slates
return preds
def kfoldCrossVal(n, slate_list, k=5, seeds = [42, 43], out_folder = '.', slate_size = None, model='rumrunner'):
os.makedirs(out_folder, exist_ok = True)
start_time = time.time()
total_iterations = len(seeds) * k
it = 0
all_errors = []
for seed in seeds:
rnd = random.Random(seed)
rnd.shuffle(slate_list)
for iter in range(k):
it += 1
print(f'seed: {seed} - iter: {iter} --- {it}/{total_iterations}')
tt = time.time()
train_list = []
test_list = []
for i in range(len(slate_list)):
if i%k == iter: #test
test_list.append(slate_list[i])
else: #train
train_list.append(slate_list[i])
test_probs = get_probs_from_list(test_list)
S_test = set()
for key in test_probs:
S_test.add(key[0])
if model in ['rumrunner', 'trainmatrix']:
train_probs = get_probs_from_list(train_list)
if model == 'rumrunner':
S_train = set()
for key in train_probs:
S_train.add(key[0])
rum = rumrunner.rumrunner_(n, train_probs, S_train, seed=seed, load_checkpoint=None, \
checkpoint_path=f'{out_folder}/seed-{seed}_iter-{iter}_rumrunner', epsilon_D_tolerance=0.00001, convergence_count=20, \
epsilon_convergence=0.0001, slate_size=slate_size, max_iterations=250, verbose=False, checkpoint_iter=100)
test_preds = slate_probabilities_from_rum(S_test, rum)
elif model == 'MNL':
mnl = train_MNL(n, train_list)
test_preds = pred_from_MNL(mnl, S_test, n)
elif model == 'MNLpcmc':
mnl = train_MNLpcmc(n, get_sums_from_list(train_list))
test_preds = pred_from_MNLpcmc(mnl, S_test, n)
elif model == 'trainmatrix':
test_preds = predict_via_training_data(train_probs, S_test, slate_size)
else:
assert False
#error = avg_l1_error(len(S_test), test_probs, test_preds)
error = RMSE(test_preds, test_probs, len(S_test))
all_errors.append(error)
print('iteration error: ', error)
print(f'iteration took {(time.time() - tt):.2f}s')
with open(f'{out_folder}/seed-{seed}_iter-{iter}_result', 'w') as f:
json.dump({'error': error, 'time': time.time()-start_time, 'model': model}, f)
stdev = np.std(np.array(all_errors))
avg = np.mean(np.array(all_errors))
print('average error: ', avg)
print('stddev: ', stdev)
print('total time: ', time.time() - start_time)
#old mse, only for 2-slates
#def rmse_2slates(D1, D2, n):
# keys = ((i,j) for i in range(n) for j in range(i+1, n))
# assert D1.keys() == D2.keys()
# toterr = 0
# for k in keys:
# if (k, k[1]) in D1: toterr += (D1[(k, k[1])] - D2[(k, k[1])])**2
# return toterr**0.5 / n
if __name__ == '__main__':
if len(sys.argv) == 1:
datasets = ['sushiA', 'SFwork', 'SFshop', 'election/a9', 'election/a17', 'election/a48', 'election/a81']
slate_sizes = [2, 3, 4, 5]
elif len(sys.argv) == 3:
datasets = [sys.argv[1]]
slate_sizes = [int(sys.argv[2])]
else:
print('python predictions.py [dataset_name] [slate_size]')
assert False
os.makedirs('./predictions/election', exist_ok=True)
for slate_size in slate_sizes:
for ds_nm in datasets:
print(ds_nm, slate_size)
ds = Dataset(f'./data/clean/{ds_nm}_{slate_size}slates', keep_sum=True)
kfoldCrossVal(ds.n, list_from_sums(ds.SUM_DSI), 5, seeds=[42+i for i in range(10)], out_folder=f'./predictions/{ds_nm}_{slate_size}slates_rumrunner', slate_size=slate_size, model='rumrunner')
kfoldCrossVal(ds.n, list_from_sums(ds.SUM_DSI), 5, seeds=[42+i for i in range(10)], out_folder=f'./predictions/{ds_nm}_{slate_size}slates_MNL', model='MNL')
kfoldCrossVal(ds.n, list_from_sums(ds.SUM_DSI), 5, seeds=[42+i for i in range(10)], out_folder=f'./predictions/{ds_nm}_{slate_size}slates_MNL_pcmc', model='MNLpcmc')
kfoldCrossVal(ds.n, list_from_sums(ds.SUM_DSI), 5, seeds=[42+i for i in range(10)], out_folder=f'./predictions/{ds_nm}_{slate_size}slates_trainmatrix', slate_size=slate_size, model='trainmatrix')
#recap evaluation
for slate_size in slate_sizes:
for ds_nm in datasets:
#print(f'==== {ds_nm} --- {slate_size}')
print(f'train tensor, {ds_nm} {slate_size}-slates: ', get_statistics(f'./predictions/{ds_nm}_{slate_size}slates_trainmatrix'))
print(f'rumrunner, {ds_nm} {slate_size}-slates: ', get_statistics(f'./predictions/{ds_nm}_{slate_size}slates_rumrunner'))
print(f'MNL, {ds_nm} {slate_size}-slates: ', get_statistics(f'./predictions/{ds_nm}_{slate_size}slates_MNL'), '\n')
print(f'MNL pcmc, {ds_nm} {slate_size}-slates: ', get_statistics(f'./predictions/{ds_nm}_{slate_size}slates_MNL_pcmc'), '\n')