Spaces:
Build error
Build error
File size: 10,344 Bytes
a446b0b |
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 |
import comet.utils.utils as utils
import comet.src.data.utils as data_utils
import comet.src.data.config as cfg
import pandas
import json
import random
import math
import torch
from tqdm import tqdm
def map_name(name):
if name == "train":
return "trn"
elif name == "test":
return "tst"
else:
return "dev"
class DataLoader(object):
def __init__(self, opt):
self.data = {}
self.data["train"] = {}
self.data["dev"] = {}
self.data["test"] = {}
self.sequences = {}
self.sequences["train"] = {}
self.sequences["dev"] = {}
self.sequences["test"] = {}
self.masks = {}
self.masks["train"] = {}
self.masks["dev"] = {}
self.masks["test"] = {}
self.offsets = {}
self.offsets["train"] = {}
self.offsets["dev"] = {}
self.offsets["test"] = {}
def offset_summary(self, split):
return self.offsets[split]["total"]
def do_take_partial_dataset(data_opts):
if data_opts.get("kr", None) is None:
return False
if data_opts.kr == 1:
return False
return True
def select_partial_dataset(data_opts, data):
num_selections = math.ceil(data_opts.kr * len(data))
return random.sample(data, num_selections)
class GenerationDataLoader(DataLoader):
def __init__(self, opt, categories):
super(GenerationDataLoader, self).__init__(opt)
self.categories = categories
self.opt = opt
for split in self.data:
self.data[split] = {"total": []}
self.offsets[split] = {"total": 0}
self.vocab_encoder = None
self.vocab_decoder = None
self.special_chars = None
self.max_event = None
self.max_effect = None
def load_data(self, path):
if ".pickle" in path:
print("Loading data from: {}".format(path))
data_utils.load_existing_data_loader(self, path)
return True
for split in self.data:
file_name = "v4_atomic_{}.csv".format(map_name(split))
df = pandas.read_csv("{}/{}".format(path, file_name), index_col=0)
df.iloc[:, :9] = df.iloc[:, :9].apply(
lambda col: col.apply(json.loads))
for cat in self.categories:
attr = df[cat]
self.data[split]["total"] += utils.zipped_flatten(zip(
attr.index, ["<{}>".format(cat)] * len(attr), attr.values))
if do_take_partial_dataset(self.opt.data):
self.data["train"]["total"] = select_partial_dataset(
self.opt.data, self.data["train"]["total"])
return False
def make_tensors(self, text_encoder, special,
splits=["train", "dev", "test"], test=False):
self.vocab_encoder = text_encoder.encoder
self.vocab_decoder = text_encoder.decoder
self.special_chars = special
sequences = {}
for split in splits:
sequences[split] = get_generation_sequences(
self.opt, self.data, split, text_encoder, test)
self.masks[split]["total"] = [(len(i[0]), len(i[1])) for
i in sequences[split]]
self.max_event = max([max([l[0] for l in self.masks[split]["total"]])
for split in self.masks])
self.max_effect = max([max([l[1] for l in self.masks[split]["total"]])
for split in self.masks])
print(self.max_event)
print(self.max_effect)
for split in splits:
num_elements = len(sequences[split])
self.sequences[split]["total"] = torch.LongTensor(
num_elements, self.max_event + self.max_effect).fill_(0)
for i, seq in enumerate(sequences[split]):
# print(self.sequences[split]["total"][i, :len(seq[0])].size())
# print(torch.FloatTensor(seq[0]).size())
self.sequences[split]["total"][i, :len(seq[0])] = \
torch.LongTensor(seq[0])
self.sequences[split]["total"][i, self.max_event:self.max_event + len(seq[1])] = \
torch.LongTensor(seq[1])
def sample_batch(self, split, bs, idxs=None):
offset = self.offsets[split]["total"]
batch = {}
# Decided not to reduce computation on here because it's all parallel
# anyway and we don't want to run out of memory in cases where we
# don't see the longest version quickly enough
if idxs:
seqs = self.sequences[split]["total"].index_select(
0, torch.LongTensor(idxs).to(
self.sequences[split]["total"].device))
else:
seqs = self.sequences[split]["total"][offset:offset + bs]
batch["sequences"] = seqs.to(cfg.device)
batch["attention_mask"] = make_attention_mask(seqs)
batch["loss_mask"] = make_loss_mask(
seqs, self.max_event, 1)
batch["key"] = ("total", offset, offset + bs)
offset += seqs.size(0)
self.offsets[split]["total"] = offset
if split == "train" and offset + bs > len(self.sequences[split]["total"]):
return batch, True
elif offset >= len(self.sequences[split]["total"]):
return batch, True
else:
return batch, False
def reset_offsets(self, splits=["train", "test", "dev"],
shuffle=True, keys=None):
if isinstance(splits, str):
splits = [splits]
for split in splits:
if keys is None:
keys = ["total"]
for key in keys:
self.offsets[split][key] = 0
if shuffle:
self.shuffle_sequences(split, keys)
def shuffle_sequences(self, split="train", keys=None):
if keys is None:
# print(type(self.data))
# print(type(self.data.keys()))
keys = self.data[split].keys()
for key in keys:
idxs = list(range(len(self.data[split][key])))
random.shuffle(idxs)
self.sequences[split][key] = \
self.sequences[split][key].index_select(
0, torch.LongTensor(idxs))
temp = [self.data[split][key][i] for i in idxs]
self.data[split][key] = temp
temp = [self.masks[split][key][i] for i in idxs]
self.masks[split][key] = temp
def prune_data_for_evaluation(data_loader, categories, split):
indices = []
for i, example in enumerate(data_loader.data[split]["total"]):
if example[1] in categories:
indices.append(i)
data_loader.masks[split]["total"] = [data_loader.masks[split]["total"][i]
for i in indices]
data_loader.sequences[split]["total"] = \
data_loader.sequences[split]["total"].index_select(
0, torch.LongTensor(indices))
data_loader.data[split]["total"] = [data_loader.data[split]["total"][i]
for i in indices]
def make_attention_mask(sequences):
return (sequences != 0).float().to(cfg.device)
def make_loss_mask(sequences, max_event, num_delim_tokens):
# print(num_delim_tokens)
# print(sequences.size())
mask = (sequences != 0).float()
mask[:, :max_event + num_delim_tokens] = 0
return mask[:, 1:].to(cfg.device)
def find_underscore_length(seq):
start = "_"
while start in seq:
start += "_"
return start[:-1]
def handle_underscores(suffix, text_encoder, prefix=False):
encoder = text_encoder.encoder
if prefix:
tok = "___"
else:
tok = find_underscore_length(suffix)
suffix_parts = [i.strip() for i in suffix.split("{}".format(tok))]
to_flatten = []
for i, part in enumerate(suffix_parts):
if part:
to_flatten.append(text_encoder.encode([part], verbose=False)[0])
if i != len(suffix_parts) - 1 and suffix_parts[i+1]:
to_flatten.append([encoder["<blank>"]])
else:
to_flatten.append([encoder["<blank>"]])
final_suffix = utils.flatten(to_flatten)
return final_suffix
def get_generation_sequences(opt, data, split, text_encoder, test):
sequences = []
count = 0
final_prefix = None
final_suffix = None
for prefix, category, suffix in tqdm(data[split]["total"]):
final_prefix, final_suffix = do_example(
text_encoder, prefix, suffix, True, True)
# if do_prefix:
# if "___" in prefix:
# final_prefix = handle_underscores(prefix, text_encoder, True)
# else:
# final_prefix = text_encoder.encode([prefix], verbose=False)[0]
# if do_suffix:
# if "_" in suffix:
# final_suffix = handle_underscores(suffix, text_encoder)
# else:
# final_suffix = text_encoder.encode([suffix], verbose=False)[0]
final = compile_final_sequence(
opt, final_prefix, final_suffix, category, text_encoder)
sequences.append(final)
count += 1
if count > 10 and test:
break
return sequences
def do_example(text_encoder, prefix, suffix, do_prefix, do_suffix):
final_prefix = None
final_suffix = None
if do_prefix:
if "___" in prefix:
final_prefix = handle_underscores(prefix, text_encoder, True)
else:
final_prefix = text_encoder.encode([prefix], verbose=False)[0]
if do_suffix:
if "_" in suffix:
final_suffix = handle_underscores(suffix, text_encoder)
else:
final_suffix = text_encoder.encode([suffix], verbose=False)[0]
return final_prefix, final_suffix
def compile_final_sequence(opt, final_prefix, final_suffix, category, text_encoder):
final = []
final.append(final_prefix)
final.append(
[text_encoder.encoder[category]]
+ final_suffix)
final[-1].append(text_encoder.encoder["<END>"])
return final
num_delimiter_tokens = {
"category": 1,
"hierarchy": 3,
"hierarchy+label": 4,
"category+hierarchy": 4,
"category+hierarchy+label": 5
}
|