forked from fastai/courses
-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathrossman_exp.py
164 lines (130 loc) · 5.32 KB
/
rossman_exp.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
train_ratio=0.9
use_dict=True
use_scaler=False
init_emb=False
split_contins=True
samp_size = 100000
#samp_size = 0
import math, keras, datetime, pandas as pd, numpy as np, keras.backend as K
import matplotlib.pyplot as plt, xgboost, operator, random, pickle, os
from sklearn_pandas import DataFrameMapper
from sklearn.preprocessing import LabelEncoder, Imputer, StandardScaler
from keras.models import Model
from keras.layers import merge, Input
from keras.layers.core import Dense, Activation, Reshape, Flatten, Dropout
from keras.layers.embeddings import Embedding
from keras.optimizers import Adam
from keras.layers.normalization import BatchNormalization
from keras.regularizers import l2
from keras import initializations
np.set_printoptions(4)
cfg = K.tf.ConfigProto()
cfg.gpu_options.allow_growth = True
K.set_session(K.tf.Session(config=cfg))
os.chdir('data/rossman')
cat_var_dict = {'Store': 50, 'DayOfWeek': 6, 'Year': 2, 'Month': 6,
'Day': 10, 'StateHoliday': 3, 'CompetitionMonthsOpen': 2,
'Promo2Weeks': 1, 'StoreType': 2, 'Assortment': 3, 'PromoInterval': 3,
'CompetitionOpenSinceYear': 4, 'Promo2SinceYear': 4, 'State': 6,
'Week': 2, 'Events': 4, 'Promo_fw': 1,
'Promo_bw': 1, 'StateHoliday_fw': 1,
'StateHoliday_bw': 1, 'SchoolHoliday_fw': 1,
'SchoolHoliday_bw': 1}
cats, contins= [o for n,o in np.load('vars.npz').items()]
y = np.load('deps.npz').items()[0][1]
if samp_size != 0:
np.random.seed(42)
idxs = sorted(np.random.choice(len(y), samp_size, replace=False))
cats= cats[idxs]
contins= contins[idxs]
y= y[idxs]
n=len(y)
train_size = int(n*train_ratio)
contins_trn_orig, contins_val_orig = contins[:train_size], contins[train_size:]
cats_trn, cats_val = cats[:train_size], cats[train_size:]
y_trn, y_val = y[:train_size], y[train_size:]
contin_map_fit = pickle.load(open('contin_maps.pickle', 'rb'))
cat_map_fit = pickle.load(open('cat_maps.pickle', 'rb'))
def cat_map_info(feat): return feat[0], len(feat[1].classes_)
co_enc = StandardScaler().fit(contins_trn_orig)
tf_contins_trn = co_enc.transform(contins_trn_orig)
tf_contins_val = co_enc.transform(contins_val_orig)
"""
def rmspe(y_pred, targ = y_valid_orig):
return math.sqrt(np.square((targ - y_pred)/targ).mean())
def log_max_inv(preds, mx = max_log_y): return np.exp(preds * mx)
def normalize_inv(preds): return preds * ystd + ymean
"""
def split_cols(arr): return np.hsplit(arr,arr.shape[1])
def emb_init(shape, name=None):
return initializations.uniform(shape, scale=0.6/shape[1], name=name)
def get_emb(feat):
name, c = cat_map_info(feat)
if use_dict:
c2 = cat_var_dict[name]
else:
c2 = (c+2)//3
if c2>50: c2=50
inp = Input((1,), dtype='int64', name=name+'_in')
if init_emb:
u = Flatten(name=name+'_flt')(Embedding(c, c2, input_length=1)(inp))
else:
u = Flatten(name=name+'_flt')(Embedding(c, c2, input_length=1, init=emb_init)(inp))
return inp,u
def get_contin(feat):
name = feat[0][0]
inp = Input((1,), name=name+'_in')
return inp, Dense(1, name=name+'_d')(inp)
def split_data():
if split_contins:
map_train = split_cols(cats_trn) + split_cols(contins_trn)
map_valid = split_cols(cats_val) + split_cols(contins_val)
else:
map_train = split_cols(cats_trn) + [contins_trn]
map_valid = split_cols(cats_val) + [contins_val]
return (map_train, map_valid)
def get_contin_one():
n_contin = contins_trn.shape[1]
contin_inp = Input((n_contin,), name='contin')
contin_out = BatchNormalization()(contin_inp)
return contin_inp, contin_out
def train(model, map_train, map_valid, bs=128, ne=10):
return model.fit(map_train, y_trn, batch_size=bs, nb_epoch=ne,
verbose=0, validation_data=(map_valid, y_val))
def get_model():
if split_contins:
conts = [get_contin(feat) for feat in contin_map_fit.features]
cont_out = [d for inp,d in conts]
cont_inp = [inp for inp,d in conts]
else:
contin_inp, contin_out = get_contin_one()
cont_out = [contin_out]
cont_inp = [contin_inp]
embs = [get_emb(feat) for feat in cat_map_fit.features]
x = merge([emb for inp,emb in embs] + cont_out, mode='concat')
x = Dropout(0.02)(x)
x = Dense(1000, activation='relu', init='uniform')(x)
x = Dense(500, activation='relu', init='uniform')(x)
x = Dense(1, activation='sigmoid')(x)
model = Model([inp for inp,emb in embs] + cont_inp, x)
model.compile('adam', 'mean_absolute_error')
#model.compile(Adam(), 'mse')
return model
for split_contins in [True, False]:
for use_dict in [True, False]:
for use_scaler in [True, False]:
for init_emb in [True, False]:
print ({'split_contins':split_contins, 'use_dict':use_dict,
'use_scaler':use_scaler, 'init_emb':init_emb})
if use_scaler:
contins_trn = tf_contins_trn
contins_val = tf_contins_val
else:
contins_trn = contins_trn_orig
contins_val = contins_val_orig
map_train, map_valid = split_data()
model = get_model()
hist = np.array(train(model, map_train, map_valid, 128, 10)
.history['val_loss'])
print(hist)
print(hist.min())