File size: 16,508 Bytes
ac7c391
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
#%%
import pickle as pkl
from typing import Dict, Tuple, List
import os
import numpy as np
import json
import logging
import argparse 
import math
from pprint import pprint
import pandas as pd
from collections import defaultdict
import copy
import time
from tqdm import tqdm

import torch
from torch.utils.data import DataLoader
import torch.backends.cudnn as cudnn
import torch.autograd as autograd

from model import Distmult, Complex, Conve
import utils

# from evaluation import evaluation

#%%
class Main(object):
    def __init__(self, args):
        self.args = args 
        
        self.model_name = '{0}_{1}_{2}_{3}_{4}'.format(args.model, args.embedding_dim, args.input_drop, args.hidden_drop, args.feat_drop)
        #leaving batches from the model_name since they do not depend on model_architecture 
        # also leaving kernel size and filters, siinice don't intend to change those
        self.model_path = 'saved_models/{0}_{1}.model'.format(args.data, self.model_name)
        
        self.log_path = 'logs/{0}_{1}_{2}_{3}.log'.format(args.data, self.model_name, args.epochs, args.train_batch_size)
        self.loss_path = 'losses/{0}_{1}_{2}_{3}.pickle'.format(args.data, self.model_name, args.epochs, args.train_batch_size)
        
        logging.basicConfig(format = '%(asctime)s - %(levelname)s - %(name)s -   %(message)s',
                            datefmt = '%m/%d/%Y %H:%M:%S',
                            level = logging.INFO,
                            filename = self.log_path)
        self.logger = logging.getLogger(__name__)
        self.logger.info(vars(self.args))
        self.logger.info('\n')
            
        
        self.device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
        
        self.load_data()
        self.model = self.add_model()
        self.optimizer = self.add_optimizer(self.model.parameters())
        
        if self.args.save_influence_map:
            self.logger.info('-------- Argument save_influence_map is set. Will use GR to compute and save influence maps ----------\n')
            # when we want to save influence during training
            self.args.add_reciprocals = False # to keep things simple
            # init an empty influence map
            self.influence_map = defaultdict(float)
            #self.influence_path = 'influence_maps/{0}_{1}.json'.format(args.data, self.model_name)
            self.influence_path = 'influence_maps/{0}_{1}.pickle'.format(args.data, self.model_name)
            # Initialize a copy of the model prams to track previous weights in an epoch
            self.previous_weights = [copy.deepcopy(param) for param in self.model.parameters()]
            self.logger.info('Shape for previous weights: {}, {}'.format(self.previous_weights[0].shape, self.previous_weights[1].shape))
    
    def load_data(self):
        ''' 
        Load the train, valid datasets
        '''
        data_path = os.path.join('processed_data', self.args.data)
        n_ent, n_rel, ent_to_id, rel_to_id = utils.generate_dicts(data_path)
        self.n_ent = n_ent
        self.n_rel = n_rel
        
        self.train_data = utils.load_data(os.path.join(data_path, 'all.txt'))
        # print(type(self.train_data), self.train_data.shape) #(1996432, 3)
        tmp = np.random.choice(a = self.train_data.shape[0], size = int(self.train_data.shape[0] * self.args.KG_valid_rate), replace=False)
        self.valid_data= self.train_data[tmp, :]

    
    def add_model(self):
        
        if self.args.model is None:
            model = Distmult(self.args, self.n_ent, self.n_rel)
        elif self.args.model == 'distmult':
            model = Distmult(self.args, self.n_ent, self.n_rel)
        elif self.args.model == 'complex':
            model = Complex(self.args, self.n_ent, self.n_rel)
        elif self.args.model == 'conve':
            model = Conve(self.args, self.n_ent, self.n_rel)
        else:
            self.logger.info('Unknown model: {0}', self.args.model)
            raise Exception("Unknown model!")
        model.to(self.device)
        return model
    
    def add_optimizer(self, parameters):
        return torch.optim.Adam(parameters, lr=self.args.lr, weight_decay=self.args.lr_decay)
    
    def save_model(self):
        state = {
            'state_dict': self.model.state_dict(),
            'optimizer': self.optimizer.state_dict(),
            'args': vars(self.args)
        }
        torch.save(state, self.model_path)
        self.logger.info('Saving model to {0}'.format(self.model_path))
    
    def load_model(self):
        self.logger.info('Loading saved model from {0}'.format(self.model_path))
        state = torch.load(self.model_path)
        model_params = state['state_dict']
        params = [(key, value.size(), value.numel()) for key, value in model_params.items()]
        for key, size, count in params:
            self.logger.info(key, size, count)
        self.model.load_state_dict(model_params)
        self.optimizer.load_state_dict(state['optimizer'])
    
    def lp_regularizer(self):
        # Apply p-norm regularization; assign weights to each param
        weight = self.args.reg_weight
        p = self.args.reg_norm
        
        trainable_params = [self.model.emb_e.weight, self.model.emb_rel.weight]
        norm = 0
        for i in range(len(trainable_params)):
            #norm += weight * trainable_params[i].norm(p = p)**p
            norm += weight * torch.sum( torch.abs(trainable_params[i]) ** p)
            
        return norm
        
    def n3_regularizer(self, factors):
        # factors are the embeddings for lhs, rel, rhs for triples in a batch
        weight = self.args.reg_weight
        p = self.args.reg_norm
        
        norm = 0
        for f in factors:
            norm += weight * torch.sum(torch.abs(f) ** p)
            
        return norm / factors[0].shape[0] # scale by number of triples in batch
    
    def get_influence_map(self):
        """
        Turns the influence map into a list, ready to be written to disc. (before: numpy)
        :return: the influence map with lists as values
        """
        assert self.args.save_influence_map == True
        
        for key in self.influence_map:
            self.influence_map[key] = self.influence_map[key].tolist()
        #self.logger.info('get_influence_map passed')
        return self.influence_map

    def evaluate(self, split, batch_size, epoch):
        """
        The same as self.run_epoch()
        """

        self.model.eval()
        losses = []

        with torch.no_grad():
            input_data = torch.from_numpy(self.valid_data.astype('int64'))
            actual_examples = input_data[torch.randperm(input_data.shape[0]), :]
            del input_data

            batch_size = self.args.valid_batch_size
            for b_begin in tqdm(range(0, actual_examples.shape[0], batch_size)):

                input_batch = actual_examples[b_begin: b_begin + batch_size]
                input_batch = input_batch.to(self.device)
                
                s,r,o = input_batch[:,0], input_batch[:,1], input_batch[:,2]
                
                emb_s = self.model.emb_e(s).squeeze(dim=1)
                emb_r = self.model.emb_rel(r).squeeze(dim=1)
                emb_o = self.model.emb_e(o).squeeze(dim=1)
                
                if self.args.add_reciprocals:
                    r_rev = r + self.n_rel
                    emb_rrev = self.model.emb_rel(r_rev).squeeze(dim=1)
                else:
                    r_rev = r
                    emb_rrev = emb_r
                    
                pred_sr = self.model.forward(emb_s, emb_r, mode='rhs')
                loss_sr = self.model.loss(pred_sr, o) # cross entropy loss
                
                pred_or = self.model.forward(emb_o, emb_rrev, mode='lhs')
                loss_or = self.model.loss(pred_or, s)
                
                total_loss = loss_sr + loss_or
                
                if (self.args.reg_weight != 0.0 and self.args.reg_norm == 3):
                    #self.logger.info('Computing regularizer weight')
                    if self.args.model == 'complex':
                        emb_dim = self.args.embedding_dim #int(self.args.embedding_dim/2)
                        lhs = (emb_s[:, :emb_dim], emb_s[:, emb_dim:])
                        rel = (emb_r[:, :emb_dim], emb_r[:, emb_dim:])
                        rel_rev = (emb_rrev[:, :emb_dim], emb_rrev[:, emb_dim:])
                        rhs = (emb_o[:, :emb_dim], emb_o[:, emb_dim:])
                        
                        #print(lhs[0].shape, lhs[1].shape)
                        factors_sr = (torch.sqrt(lhs[0] ** 2 + lhs[1] ** 2),
                                    torch.sqrt(rel[0] ** 2 + rel[1] ** 2),
                                    torch.sqrt(rhs[0] ** 2 + rhs[1] ** 2)
                                )
                        factors_or = (torch.sqrt(lhs[0] ** 2 + lhs[1] ** 2),
                                    torch.sqrt(rel_rev[0] ** 2 + rel_rev[1] ** 2),
                                    torch.sqrt(rhs[0] ** 2 + rhs[1] ** 2)
                                )
                    else:
                        factors_sr = (emb_s, emb_r, emb_o)
                        factors_or = (emb_s, emb_rrev, emb_o)
                        
                    total_loss  += self.n3_regularizer(factors_sr)
                    total_loss  += self.n3_regularizer(factors_or)
                    
                if (self.args.reg_weight != 0.0 and self.args.reg_norm == 2):
                    total_loss += self.lp_regularizer()

                losses.append(total_loss.item())

        loss = np.mean(losses)
        self.logger.info('[Epoch:{}]:  Validating Loss:{:.6}\n'.format(epoch, loss))
        return loss


    def run_epoch(self, epoch):
        self.model.train()
        losses = []
        
        #shuffle the train dataset
        input_data = torch.from_numpy(self.train_data.astype('int64'))
        actual_examples = input_data[torch.randperm(input_data.shape[0]), :]
        del input_data
        
        batch_size = self.args.train_batch_size
        
        for b_begin in tqdm(range(0, actual_examples.shape[0], batch_size)):
            self.optimizer.zero_grad()
            input_batch = actual_examples[b_begin: b_begin + batch_size]
            input_batch = input_batch.to(self.device)
            
            s,r,o = input_batch[:,0], input_batch[:,1], input_batch[:,2]
            
            emb_s = self.model.emb_e(s).squeeze(dim=1)
            emb_r = self.model.emb_rel(r).squeeze(dim=1)
            emb_o = self.model.emb_e(o).squeeze(dim=1)
            
            if self.args.add_reciprocals:
                r_rev = r + self.n_rel
                emb_rrev = self.model.emb_rel(r_rev).squeeze(dim=1)
            else:
                r_rev = r
                emb_rrev = emb_r
                
            pred_sr = self.model.forward(emb_s, emb_r, mode='rhs')
            loss_sr = self.model.loss(pred_sr, o) # loss is cross entropy loss
            
            pred_or = self.model.forward(emb_o, emb_rrev, mode='lhs')
            loss_or = self.model.loss(pred_or, s)
            
            total_loss = loss_sr + loss_or
            
            if (self.args.reg_weight != 0.0 and self.args.reg_norm == 3):
                #self.logger.info('Computing regularizer weight')
                if self.args.model == 'complex':
                    emb_dim = self.args.embedding_dim #int(self.args.embedding_dim/2)
                    lhs = (emb_s[:, :emb_dim], emb_s[:, emb_dim:])
                    rel = (emb_r[:, :emb_dim], emb_r[:, emb_dim:])
                    rel_rev = (emb_rrev[:, :emb_dim], emb_rrev[:, emb_dim:])
                    rhs = (emb_o[:, :emb_dim], emb_o[:, emb_dim:])
                    
                    #print(lhs[0].shape, lhs[1].shape)
                    factors_sr = (torch.sqrt(lhs[0] ** 2 + lhs[1] ** 2),
                                torch.sqrt(rel[0] ** 2 + rel[1] ** 2),
                                torch.sqrt(rhs[0] ** 2 + rhs[1] ** 2))
                    factors_or = (torch.sqrt(lhs[0] ** 2 + lhs[1] ** 2),
                                torch.sqrt(rel_rev[0] ** 2 + rel_rev[1] ** 2),
                                torch.sqrt(rhs[0] ** 2 + rhs[1] ** 2))
                else:
                    factors_sr = (emb_s, emb_r, emb_o)
                    factors_or = (emb_s, emb_rrev, emb_o)
                    
                total_loss  += self.n3_regularizer(factors_sr)
                total_loss  += self.n3_regularizer(factors_or)
                
            if (self.args.reg_weight != 0.0 and self.args.reg_norm == 2):
                total_loss += self.lp_regularizer()
                    
                    
            total_loss.backward()
            self.optimizer.step()
            losses.append(total_loss.item())
            
            if self.args.save_influence_map: #for gradient rollback
                with torch.no_grad():
                    prev_emb_e = self.previous_weights[0]
                    prev_emb_rel = self.previous_weights[1]
                    # need to compute the influence value per-triple
                    for idx in range(input_batch.shape[0]):
                        head, rel, tail = s[idx], r[idx], o[idx]
                        inf_head = (emb_s[idx] - prev_emb_e[head]).cpu().detach().numpy()
                        inf_tail = (emb_o[idx] - prev_emb_e[tail]).cpu().detach().numpy()
                        inf_rel = (emb_r[idx] - prev_emb_rel[rel]).cpu().detach().numpy()
                        #print(inf_head.shape, inf_tail.shape, inf_rel.shape)

                        #write the influences to dictionary
                        key_trip = '{0}_{1}_{2}'.format(head.item(), rel.item(), tail.item())
                        key = '{0}_s'.format(key_trip)
                        self.influence_map[key] += inf_head
                        #self.logger.info('Written to influence map. Key: {0}, Value shape: {1}'.format(key, inf_head.shape))
                        key = '{0}_r'.format(key_trip)
                        self.influence_map[key] += inf_rel
                        key = '{0}_o'.format(key_trip)
                        self.influence_map[key] += inf_tail

                    # update the previous weights to be tracked
                    self.previous_weights = [copy.deepcopy(param) for param in self.model.parameters()]
            
            if (b_begin%5000 == 0) or (b_begin== (actual_examples.shape[0]-1)):
                self.logger.info('[E:{} | {}]: Train Loss:{:.6}'.format(epoch, b_begin, np.mean(losses)))
                
        loss = np.mean(losses)
        self.logger.info('[Epoch:{}]:  Training Loss:{:.6}\n'.format(epoch, loss))
        return loss
    
    def fit(self):
        self.model.init()
        self.logger.info(self.model)
        
        self.logger.info('------ Start the model training ------')
        start_time = time.time()
        self.logger.info('Start time: {0}'.format(str(start_time)))
    
    
        train_losses = []
        valid_losses = []
        best_val = 10000000000.
        for epoch in range(self.args.epochs):

            print("="*15,'epoch:',epoch,'='*15)
            train_loss = self.run_epoch(epoch)
            train_losses.append(train_loss)

            if train_loss < best_val:
                best_val = train_loss
                self.save_model()
            print("Train loss: {0}, Best loss: {1}\n\n".format(train_loss, best_val))

                
        with open(self.loss_path, "wb") as fl: 
            pkl.dump({"train loss":train_losses, "valid loss":valid_losses}, fl)
        self.logger.info('Time taken to train the model: {0}'.format(str(time.time() - start_time)))
        start_time = time.time()
        
        if self.args.save_influence_map: #save the influence map
            with open(self.influence_path, "wb") as fl:   #Pickling
                pkl.dump(self.get_influence_map(), fl)
            self.logger.info('Finished saving influence map')
            self.logger.info('Time taken to save the influence map: {0}'.format(str(time.time() - start_time)))
    
#%%
parser = utils.get_argument_parser()

args = parser.parse_args()
args = utils.set_hyperparams(args)

utils.seed_all(args.seed)
np.set_printoptions(precision=5)
cudnn.benchmark = False

model = Main(args)
model.fit()