-
Notifications
You must be signed in to change notification settings - Fork 10
/
data_loader.py
534 lines (372 loc) · 21.2 KB
/
data_loader.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
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
import os
import numpy as np
import pandas as pd
import time
import sys
import math
import pickle
from sklearn.utils import shuffle
import preprocess
def run_cluster_calculate_norm_stats():
H5_PATH_PREFIX = "/users/ak1774/scratch/esport/death_prediction/cluster_scripts/parse_job_out/parsed_files/"
H5_FILE_LIST = "/users/ak1774/scratch/esport/death_prediction/all_h5_files.txt"
now = time.time()
h5_files = get_h5_file_list(H5_PATH_PREFIX,H5_FILE_LIST)
h5_files = np.random.choice(h5_files,150)
data = load_data_chunk(h5_files,worker_id=None,num_workers=None)
#data = load_all_data(H5_PATH_PREFIX,H5_FILE_LIST)
print("Loading took: ", time.time()-now)
print("Data shape: ", data.shape)
sys.stdout.flush()
now = time.time()
norm_stats = calculate_normalization_stats(data)
print("Collecting min max took: ", time.time()-now)
sys.stdout.flush()
now = time.time()
with open("norm_stats.pickle", 'wb') as f:
pickle.dump(norm_stats, f, pickle.HIGHEST_PROTOCOL)
print("Pickleing took: ", time.time()-now)
sys.stdout.flush()
def run_cluster_randomize(data_type):
if data_type == "train":
H5_PATH_PREFIX = "/users/ak1774/scratch/esport/death_prediction/cluster_scripts/parse_job_out/parsed_files/"
H5_FILE_LIST = "/users/ak1774/scratch/esport/death_prediction/cluster_scripts/training_files.txt"
OUT_FOLDER = "/mnt/lustre/groups/cs-dclabs-2019/esport/death_prediction_data/randomized_data/train/"
elif data_type == "test":
H5_PATH_PREFIX = "/users/ak1774/scratch/esport/death_prediction/cluster_scripts/parse_job_out/parsed_files/"
H5_FILE_LIST = "/users/ak1774/scratch/esport/death_prediction/cluster_scripts/test_files.txt"
OUT_FOLDER = "/mnt/lustre/groups/cs-dclabs-2019/esport/death_prediction_data/randomized_data/test/"
elif data_type == "validation":
H5_PATH_PREFIX = "/users/ak1774/scratch/esport/death_prediction/cluster_scripts/parse_job_out/parsed_files/"
H5_FILE_LIST = "/users/ak1774/scratch/esport/death_prediction/cluster_scripts/validation_files.txt"
OUT_FOLDER = "/mnt/lustre/groups/cs-dclabs-2019/esport/death_prediction_data/randomized_data/validation/"
WORKER_ID = int(os.environ['SLURM_ARRAY_TASK_ID'])
NUM_WORKERS = int(os.environ['SLURM_ARRAY_TASK_COUNT'])
now = time.time()
h5_files = get_h5_file_list(H5_PATH_PREFIX,H5_FILE_LIST)
data = load_data_chunk(h5_files,WORKER_ID,NUM_WORKERS)
# shuffle the data
data = shuffle(data)
now = time.time()
DATA_CHUNK_SIZE = 4000
num_chunks = int(data.shape[0] / DATA_CHUNK_SIZE)
rest = data.shape[0] - num_chunks * DATA_CHUNK_SIZE
# the first rest chunk will contain 301 points, whis way every point is used, and they all have a similar size
# NOTE I assume thet a worker have at least DATA_CHUNK_SIZE*DATA_CHUNK_SIZE datapoints, otherwise this tactic can fail...
# Actually we just throw away a little bit of data, that is fine...
current_index = 0
for i in range(num_chunks):
size = DATA_CHUNK_SIZE
if i < rest:
size += 1
duta_chunk = data[current_index:current_index+size]
duta_chunk.to_hdf(OUT_FOLDER + 'data_chunk_' + str(WORKER_ID) + "_" + str(i) + '.h5', key='duta_chunk', mode='w', complevel = 9,complib='zlib')
current_index += size
print("Saving took: ", time.time()-now)
sys.stdout.flush()
def run_cluster_normalize(data_type = None):
if data_type == None:
H5_PATH_PREFIX = "/users/ak1774/scratch/esport/death_prediction/cluster_scripts/parse_job_out/parsed_files/"
H5_FILE_LIST = "/users/ak1774/scratch/esport/death_prediction/all_h5_files.txt"
OUT_FOLDER = "data_out/"
elif data_type == "train":
H5_PATH_PREFIX = "/users/ak1774/scratch/esport/death_prediction/cluster_scripts/parse_job_out/parsed_files/"
H5_FILE_LIST = "/users/ak1774/scratch/esport/death_prediction/cluster_scripts/training_files.txt"
OUT_FOLDER = "/mnt/lustre/groups/cs-dclabs-2019/esport/death_prediction_data/randomized_data/train/"
elif data_type == "test":
H5_PATH_PREFIX = "/users/ak1774/scratch/esport/death_prediction/cluster_scripts/parse_job_out/parsed_files/"
H5_FILE_LIST = "/users/ak1774/scratch/esport/death_prediction/cluster_scripts/test_files.txt"
OUT_FOLDER = "/mnt/lustre/groups/cs-dclabs-2019/esport/death_prediction_data/randomized_data/test/"
elif data_type == "validation":
H5_PATH_PREFIX = "/users/ak1774/scratch/esport/death_prediction/cluster_scripts/parse_job_out/parsed_files/"
H5_FILE_LIST = "/users/ak1774/scratch/esport/death_prediction/cluster_scripts/validation_files.txt"
OUT_FOLDER = "/mnt/lustre/groups/cs-dclabs-2019/esport/death_prediction_data/randomized_data/validation/"
WORKER_ID = int(os.environ['SLURM_ARRAY_TASK_ID'])
NUM_WORKERS = int(os.environ['SLURM_ARRAY_TASK_COUNT'])
now = time.time()
norm_stats = None
with open("norm_stats.pickle", 'rb') as f:
norm_stats = pickle.load(f)
print("Unpickleing took: ", time.time()-now)
sys.stdout.flush()
now = time.time()
h5_files = get_h5_file_list(H5_PATH_PREFIX,H5_FILE_LIST)
data_chunk = load_data_chunk(h5_files,WORKER_ID,NUM_WORKERS)
RECALCULATE_WHO_DIES_NEXT_LABELS = False
if RECALCULATE_WHO_DIES_NEXT_LABELS:
data_chunk = preprocess.create_who_dies_next_labels(data_chunk)
print("Loading took: ", time.time()-now)
print("Data shape: ", data_chunk.shape)
sys.stdout.flush()
now = time.time()
data = normalize_data(data_chunk,norm_stats)
print("Normalizing took: ", time.time()-now)
sys.stdout.flush()
# shuffle the data
data = shuffle(data)
now = time.time()
DATA_CHUNK_SIZE = 4000
num_chunks = int(data.shape[0] / DATA_CHUNK_SIZE)
rest = data.shape[0] - num_chunks * DATA_CHUNK_SIZE
# the first rest chunk will contain 301 points, whis way every point is used, and they all have a similar size
# NOTE I assume thet a worker have at least DATA_CHUNK_SIZE*DATA_CHUNK_SIZE datapoints, otherwise this tactic can fail...
# Actually we just throw away a little bit of data, that is fine...
current_index = 0
for i in range(num_chunks):
size = DATA_CHUNK_SIZE
if i < rest:
size += 1
duta_chunk = data[current_index:current_index+size]
duta_chunk.to_hdf(OUT_FOLDER + 'data_chunk_' + str(WORKER_ID) + "_" + str(i) + '.h5', key='duta_chunk', mode='w', complevel = 9,complib='zlib')
current_index += size
print("Saving took: ", time.time()-now)
sys.stdout.flush()
def get_h5_file_list(path_prefix,h5_file_list_path):
with open(h5_file_list_path) as f:
h5_files = f.readlines()
h5_files = [x.strip() for x in h5_files]
h5_files = [x.replace("./",path_prefix) for x in h5_files]
return h5_files
def load_data_from_file(filename):
return load_data_chunk([filename],worker_id=None,num_workers=None)
def load_all_data(path_prefix,h5_file_list_path):
h5_files = get_h5_file_list(path_prefix,h5_file_list_path)
return load_data_chunk(h5_files,worker_id=None,num_workers=None)
# will load all data if worker id is None
# TODO BUG: this is horribly inefficient. Concatenate will reallocate the whole dataset at every call...
# Solution: allocate a large memory beforehand... If dont know how big we need, use doubleing whenewer we run out? Maybe shrink in the end.
def load_data_chunk(h5_files,worker_id=None,num_workers=None):
if worker_id is not None:
files_per_worker = int(math.ceil(float(len(h5_files)) / num_workers))
h5_files = h5_files[worker_id*files_per_worker : (worker_id+1)*files_per_worker]
#h5_files = h5_files[0:10] # debug only take the first 10
# read in all the data, and concatenate it into on big data frame
data = None
num_has_nan = 0
for i,filename in enumerate(h5_files):
if(i % 10) == 9:
print("Loading file: ",i)
sys.stdout.flush()
#if data is not None:
# data.info()
if data is None:
data = pd.read_hdf(filename)
if data.isnull().values.any():
data = None
num_has_nan += 1
continue
else:
new_data = pd.read_hdf(filename)
if new_data.isnull().values.any():
num_has_nan += 1
continue
data = pd.concat((data,new_data),sort=False)
# print("Ratio of corrupt files: ",float(num_has_nan) / len(h5_files))
return data
def postprocess_data(data):
data = preprocess.addHeroOneHotEncoding(data)
return data
# These are called Correct because there was an incorrect version, and I did not want to break compatibility by reusing the same function name.
def getFeatureCorrectMinimal(data):
only_include_list = []
only_include_list.append("iHealth")
only_include_list.append("iTotalEarnedGold")
only_include_list.append("lifeState")
only_include_list.append("_pos_")
only_include_list.append("_proximity_")
exclude_if_contains_list = []
exclude_if_contains_list.append("ability") # to exclude ability level and stuff..
exclude_if_contains_list.append("_delta_closest_tower_distance")
exclude_if_contains_list.append("_delta_proximity_")
return getFeatureIndicies(data,exclude_if_contains_list,only_include_list)
def getFeatureCorrectMedium(data):
exclude_if_contains_list = []
exclude_if_contains_list.append("_ability_") # all ability features
exclude_if_contains_list.append("_hero_one_hot_") # hero id
return getFeatureIndicies(data,exclude_if_contains_list)
def getFeatureCorrectAll(data):
return getFeatureIndicies(data)
def getFeatureIndicies(data,exclude_if_contains_list = None,only_include_list = None):
# get an example row
example_row = data.sample(n=1,replace=False)
example_row = postprocess_data(example_row)
labels = [(i,label) for i,label in enumerate(list(example_row))]
if only_include_list is not None:
filtered_labels = []
for i,label in labels:
for include_label in only_include_list:
if include_label in label:
filtered_labels.append((i,label))
labels = filtered_labels
if exclude_if_contains_list is not None:
for exclude_pattern in exclude_if_contains_list:
labels = [(i,label) for i,label in labels if exclude_pattern not in label]
hero_feature_indicies = []
for i in range(10):
hero_labels = preprocess.select_features_of_hero(i,labels)
hero_feature_indicies.append(preprocess.labels_to_indicies(hero_labels))
hero_feature_indicies[-1].append(0) # dont forget the time
return hero_feature_indicies
def getLableIndiciesWhoDiesNext(data):
example_row = data.sample(n=1,replace=False)
example_row = postprocess_data(example_row)
labels = [(i,label) for i,label in enumerate(list(example_row))]
classification_label = preprocess.labels_to_indicies(preprocess.select_features_by_name("who_dies_next",labels))
return classification_label
def getLabelIndicies_die_in_n(data,label_name):
example_row = data.sample(n=1,replace=False)
example_row = postprocess_data(example_row)
labels = [(i,label) for i,label in enumerate(list(example_row))]
classification_label = preprocess.labels_to_indicies(preprocess.select_features_by_name(label_name,labels))
return classification_label
def getLabelIndicies_die_in_5(data):
example_row = data.sample(n=1,replace=False)
example_row = postprocess_data(example_row)
labels = [(i,label) for i,label in enumerate(list(example_row))]
classification_label = preprocess.labels_to_indicies(preprocess.select_features_by_name("die_in_5",labels))
return classification_label
def getLabelIndicies_die_in_10(data):
example_row = data.sample(n=1,replace=False)
example_row = postprocess_data(example_row)
labels = [(i,label) for i,label in enumerate(list(example_row))]
classification_label = preprocess.labels_to_indicies(preprocess.select_features_by_name("die_in_10",labels))
return classification_label
def getLabelIndicies_die_in_15(data):
example_row = data.sample(n=1,replace=False)
example_row = postprocess_data(example_row)
labels = [(i,label) for i,label in enumerate(list(example_row))]
classification_label = preprocess.labels_to_indicies(preprocess.select_features_by_name("die_in_15",labels))
return classification_label
def getLabelIndicies_die_in_20(data):
example_row = data.sample(n=1,replace=False)
example_row = postprocess_data(example_row)
labels = [(i,label) for i,label in enumerate(list(example_row))]
classification_label = preprocess.labels_to_indicies(preprocess.select_features_by_name("die_in_20",labels))
return classification_label
# usage:
# small datasets:
# data = normalize_data(data)
# large datasets:
# stats = calculate_normalization_stats(data)
# data_chunk = normalize_data(data_chunk,stats)
def normalize_data(data,normalization_stats = None):
if normalization_stats is None:
normalization_stats = calculate_normalization_stats(data)
for feature_index,(label,min_value,max_value) in enumerate(normalization_stats):
for hero_i in range(10):
true_label = "player_" + str(hero_i) + label
true_label = true_label.replace("TEAM_SLOT_IDX","000" + str(hero_i % 5))
true_label = true_label.replace("PLAYER_IDX","000" + str(hero_i))
if (max_value - min_value) == 0: # does not change, drop it
data = data.drop(true_label,axis=1)
if hero_i == 0:
print(true_label," is useless!!! It is dropped")
else:
# kwargs is weird, if I want to pass the value of the string reather than the name, i must use the dictionary syntax...
# reather than this: data = data.assign(true_label=(data[true_label] - min_value) / (max_value - min_value))
data = data.assign(**{true_label : (data[true_label] - min_value) / (max_value - min_value)})
return data
def calculate_normalization_stats(data):
# NOTE there are also "label_" and stat_" labels (eg stat_0_time_until_next_death ) which we dont normalize
# we dont use them as features
# normalization takes too long, so we calculate min and max based on a fraction of the data
representative_sample_size = min(10000,data.shape[0])
take_every_n_th = int(math.floor(float(data.shape[0]) / representative_sample_size))
# normalize time
max_value = data["time"].max()
min_value = data["time"].min()
data = data.assign(time=(data["time"] - min_value) / (max_value - min_value))
# normalize hero features
# hero features are the same for every hero
# we need to get the max and min looking at all the hero slots
# dont normalize m_nSelectedHeroID, it is going to be turend into one hot encoding
labels = [(i,label) for i,label in enumerate(list(data))]
hero_labels = [label for i,label in preprocess.select_features_of_hero(0,labels) if "m_nSelectedHeroID" not in label ]
hero_labels = [label.replace("player_0","") for label in hero_labels]
hero_labels = [label.replace("0000","TEAM_SLOT_IDX") if ("m_vecDataTeam" in label) else label for label in hero_labels]
hero_labels = [label.replace("0000","PLAYER_IDX") if ("m_vecPlayerTeamData" in label) else label for label in hero_labels]
normalization_stats = []
for label_i,label in enumerate(hero_labels):
max_value = np.finfo(np.float32).min
min_value = np.finfo(np.float32).max
for hero_i in range(10):
true_label = "player_" + str(hero_i) + label
true_label = true_label.replace("TEAM_SLOT_IDX","000" + str(hero_i % 5))
true_label = true_label.replace("PLAYER_IDX","000" + str(hero_i))
max_value = max(max_value,data[true_label][::take_every_n_th].max())
min_value = min(min_value,data[true_label][::take_every_n_th].min())
# TODO BUG, this normalize excpects min max order, it does normalize kind of correctly: (val-max) / -range, (max becomes 0, min becomes 1)
# Since it does not matter for neural networks, it was left unfixed.
# when fixing it in the future: fic here, in normalize() and in test.py
normalization_stats.append((label,max_value,min_value))
return normalization_stats
# just get a random sample
# we dont care about getting balanced labels
def getBatchNaive(data,batch_size,hero_feature_indicies,classification_labels):
data_batch = data.sample(n=batch_size,replace=False)
# this is done only now, because it would takes up too much memory
data_batch = postprocess_data(data_batch)
num_features_per_hero = len(hero_feature_indicies[0])
num_features_total = num_features_per_hero * 10
#hero_features = np.zeros((batch_size,num_features_total))
hero_features = []
for i in range(10):
#hero_features[:,(num_features_per_hero*i):(num_features_per_hero*(i+1))] = data_batch.values[:,hero_feature_indicies[i]]
hero_features.append(data_batch.values[:,hero_feature_indicies[i]].astype(np.float32))
classification_label_values = data_batch.values[:,classification_labels].astype(np.float32)
return hero_features,classification_label_values
def getSequencialNaive(data,hero_feature_indicies,classification_labels):
data_batch = data#data.sample(n=10000,replace=False)
# this is done only now, because it would takes up too much memory
data_batch = postprocess_data(data_batch)
hero_features = []
for i in range(10):
#hero_features[:,(num_features_per_hero*i):(num_features_per_hero*(i+1))] = data_batch.values[:,hero_feature_indicies[i]]
hero_features.append(data_batch.values[:,hero_feature_indicies[i]].astype(np.float32))
classification_label_values = data_batch.values[:,classification_labels].astype(np.float32)
return hero_features,classification_label_values
def getBalancedBatchForPlayer(data,player_i,batch_size,hero_feature_indicies,classification_labels,get_death_times=False):
# get a batch, where half of the time the selected player dies, the other half not
#player_dies_mask = data["label_who_dies_next_" + str(player_i)].values > 0.5
# classification label indicies is contains the indicies of labels like "player_0_die_in_10"
player_dies_mask = data.values[:,classification_labels[player_i]] > 0.5
num_sample_from_die = int(batch_size/2)
num_sample_from_not_die = batch_size - num_sample_from_die
have_enough_unique_data = sum(player_dies_mask) > num_sample_from_die
data_batch_die = data[player_dies_mask].sample(n=num_sample_from_die,replace=(have_enough_unique_data == False))
have_enough_unique_data = sum(~player_dies_mask) > num_sample_from_not_die
data_batch_not_die = data[~player_dies_mask].sample(n=num_sample_from_not_die,replace=(have_enough_unique_data == False))
data_batch = pd.concat([data_batch_die,data_batch_not_die])
# this is done only now, because it would takes up too much memory
data_batch = postprocess_data(data_batch)
hero_features = []
for i in range(10):
hero_features.append(data_batch.values[:,hero_feature_indicies[i]].astype(np.float32))
classification_label_values = data_batch.values[:,classification_labels].astype(np.float32)
if get_death_times == True:
labels = [(i,label) for i,label in enumerate(list(data))]
death_time_indicies = preprocess.labels_to_indicies(preprocess.select_features_by_name("time_until_next_death",labels))
death_times = data_batch.values[:,death_time_indicies]
return hero_features,classification_label_values,death_times
return hero_features,classification_label_values
def getBatchBalanced(data,batch_size,hero_feature_indicies,classification_labels,get_death_times=False):
no_one_dies_mask = data["label_who_dies_next_10"].values > 0.5
num_sample_from_die = int(batch_size * 10.0/11)
num_sample_from_not_die = batch_size - num_sample_from_die
have_enough_unique_data = sum(~no_one_dies_mask) > num_sample_from_die
data_batch_die = data[~no_one_dies_mask].sample(n=num_sample_from_die,replace=(have_enough_unique_data == False))
have_enough_unique_data = sum(no_one_dies_mask) > num_sample_from_not_die
data_batch_not_die = data[no_one_dies_mask].sample(n=num_sample_from_not_die,replace=(have_enough_unique_data == False))
data_batch = pd.concat([data_batch_die,data_batch_not_die])
# this is done only now, because it would takes up too much memory
data_batch = postprocess_data(data_batch)
hero_features = []
for i in range(10):
hero_features.append(data_batch.values[:,hero_feature_indicies[i]].astype(np.float32))
classification_label_values = data_batch.values[:,classification_labels].astype(np.float32)
if get_death_times == True:
labels = [(i,label) for i,label in enumerate(list(data))]
death_time_indicies = preprocess.labels_to_indicies(preprocess.select_features_by_name("time_until_next_death",labels))
death_times = data_batch.values[:,death_time_indicies]
return hero_features,classification_label_values,death_times
return hero_features,classification_label_values