Hugging Face's logo

nlp-tlp
/
mwo-ner-test

Token Classification
Flair
PyTorch
English
sequence-tagger-model
Model card Files Community
Michael Stewart commited on
Commit
2a843a2
·
1 Parent(s): c23b643

initial commit

Browse files
Files changed (5) hide show
  1. .gitignore +1 -0
  2. README.md +3 -0
  3. loss.tsv +11 -0
  4. pytorch_model.bin +3 -0
  5. training.log +178 -0
.gitignore ADDED
@@ -0,0 +1 @@
 
 
1
+ final-model.pt
README.md ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ ## MWO NER Test
2
+
3
+ A flair-based NER model for MWOs. This is just a test (trained on a small sample of annotated MWOs) to try out Huggingface - a full model will be posted later.
loss.tsv ADDED
@@ -0,0 +1,11 @@
 
 
 
 
 
 
 
 
 
 
 
 
1
+ EPOCH TIMESTAMP BAD_EPOCHS LEARNING_RATE TRAIN_LOSS DEV_LOSS DEV_PRECISION DEV_RECALL DEV_F1 DEV_ACCURACY
2
+ 1 06:06:10 0 0.1000 3.730552109629519 3.0921895503997803 0.2706 0.3594 0.3087 0.2396
3
+ 2 06:06:12 1 0.1000 2.8029938053022994 2.827807903289795 0.2464 0.2656 0.2556 0.1848
4
+ 3 06:06:13 0 0.1000 2.538649709601151 2.7169787883758545 0.4324 0.25 0.3168 0.2162
5
+ 4 06:06:14 1 0.1000 2.3749914594024784 2.749277114868164 0.0 0.0 0.0 0.0
6
+ 5 06:06:15 2 0.1000 2.2541055717931586 2.7171242237091064 0.3 0.2812 0.2903 0.2
7
+ 6 06:06:15 3 0.1000 2.183038889155214 2.6913464069366455 0.2456 0.2188 0.2314 0.1522
8
+ 7 06:06:15 0 0.1000 2.0639273608744384 2.2334768772125244 0.5 0.5 0.5 0.3902
9
+ 8 06:06:17 1 0.1000 1.9046660666523674 2.3985493183135986 0.3846 0.3125 0.3448 0.2247
10
+ 9 06:06:17 2 0.1000 1.8903841644163557 2.121741533279419 0.4429 0.4844 0.4627 0.3483
11
+ 10 06:06:17 3 0.1000 1.755499727812856 2.2982170581817627 0.38 0.2969 0.3333 0.2135
pytorch_model.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:7d80af5e5f2519fefdf850fd4d622c565cce4f7629f1d38cdc9ff6be166b5a34
3
+ size 394226083
training.log ADDED
@@ -0,0 +1,178 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ 2022-08-25 06:06:10,564 ----------------------------------------------------------------------------------------------------
2
+ 2022-08-25 06:06:10,565 Model: "SequenceTagger(
3
+ (embeddings): StackedEmbeddings(
4
+ (list_embedding_0): FlairEmbeddings(
5
+ (lm): LanguageModel(
6
+ (drop): Dropout(p=0.25, inplace=False)
7
+ (encoder): Embedding(275, 100)
8
+ (rnn): LSTM(100, 2048)
9
+ (decoder): Linear(in_features=2048, out_features=275, bias=True)
10
+ )
11
+ )
12
+ (list_embedding_1): FlairEmbeddings(
13
+ (lm): LanguageModel(
14
+ (drop): Dropout(p=0.25, inplace=False)
15
+ (encoder): Embedding(275, 100)
16
+ (rnn): LSTM(100, 2048)
17
+ (decoder): Linear(in_features=2048, out_features=275, bias=True)
18
+ )
19
+ )
20
+ )
21
+ (word_dropout): WordDropout(p=0.05)
22
+ (locked_dropout): LockedDropout(p=0.5)
23
+ (embedding2nn): Linear(in_features=4096, out_features=4096, bias=True)
24
+ (rnn): LSTM(4096, 256, batch_first=True, bidirectional=True)
25
+ (linear): Linear(in_features=512, out_features=47, bias=True)
26
+ (loss_function): ViterbiLoss()
27
+ (crf): CRF()
28
+ )"
29
+ 2022-08-25 06:06:10,567 ----------------------------------------------------------------------------------------------------
30
+ 2022-08-25 06:06:10,567 Corpus: "Corpus: 100 train + 20 dev + 20 test sentences"
31
+ 2022-08-25 06:06:10,567 ----------------------------------------------------------------------------------------------------
32
+ 2022-08-25 06:06:10,568 Parameters:
33
+ 2022-08-25 06:06:10,568 - learning_rate: "0.100000"
34
+ 2022-08-25 06:06:10,568 - mini_batch_size: "32"
35
+ 2022-08-25 06:06:10,568 - patience: "3"
36
+ 2022-08-25 06:06:10,569 - anneal_factor: "0.5"
37
+ 2022-08-25 06:06:10,569 - max_epochs: "10"
38
+ 2022-08-25 06:06:10,569 - shuffle: "True"
39
+ 2022-08-25 06:06:10,569 - train_with_dev: "False"
40
+ 2022-08-25 06:06:10,570 - batch_growth_annealing: "False"
41
+ 2022-08-25 06:06:10,570 ----------------------------------------------------------------------------------------------------
42
+ 2022-08-25 06:06:10,571 Model training base path: "db/kgc_models/test-kgc-sm-flair/trained_model"
43
+ 2022-08-25 06:06:10,571 ----------------------------------------------------------------------------------------------------
44
+ 2022-08-25 06:06:10,571 Device: cuda:0
45
+ 2022-08-25 06:06:10,571 ----------------------------------------------------------------------------------------------------
46
+ 2022-08-25 06:06:10,572 Embeddings storage mode: gpu
47
+ 2022-08-25 06:06:10,572 ----------------------------------------------------------------------------------------------------
48
+ 2022-08-25 06:06:10,659 epoch 1 - iter 1/4 - loss 4.03916875 - samples/sec: 371.51 - lr: 0.100000
49
+ 2022-08-25 06:06:10,731 epoch 1 - iter 2/4 - loss 3.89573674 - samples/sec: 443.37 - lr: 0.100000
50
+ 2022-08-25 06:06:10,805 epoch 1 - iter 3/4 - loss 3.73763981 - samples/sec: 441.41 - lr: 0.100000
51
+ 2022-08-25 06:06:10,853 epoch 1 - iter 4/4 - loss 3.73055211 - samples/sec: 669.71 - lr: 0.100000
52
+ 2022-08-25 06:06:10,854 ----------------------------------------------------------------------------------------------------
53
+ 2022-08-25 06:06:10,854 EPOCH 1 done: loss 3.7306 - lr 0.100000
54
+ 2022-08-25 06:06:10,934 Evaluating as a multi-label problem: False
55
+ 2022-08-25 06:06:10,941 DEV : loss 3.0921895503997803 - f1-score (micro avg) 0.3087
56
+ 2022-08-25 06:06:10,942 BAD EPOCHS (no improvement): 0
57
+ 2022-08-25 06:06:10,944 saving best model
58
+ 2022-08-25 06:06:12,789 ----------------------------------------------------------------------------------------------------
59
+ 2022-08-25 06:06:12,816 epoch 2 - iter 1/4 - loss 2.91073551 - samples/sec: 1237.62 - lr: 0.100000
60
+ 2022-08-25 06:06:12,842 epoch 2 - iter 2/4 - loss 2.94079960 - samples/sec: 1251.27 - lr: 0.100000
61
+ 2022-08-25 06:06:12,874 epoch 2 - iter 3/4 - loss 2.78518824 - samples/sec: 1014.72 - lr: 0.100000
62
+ 2022-08-25 06:06:12,891 epoch 2 - iter 4/4 - loss 2.80299381 - samples/sec: 1978.97 - lr: 0.100000
63
+ 2022-08-25 06:06:12,892 ----------------------------------------------------------------------------------------------------
64
+ 2022-08-25 06:06:12,892 EPOCH 2 done: loss 2.8030 - lr 0.100000
65
+ 2022-08-25 06:06:12,932 Evaluating as a multi-label problem: False
66
+ 2022-08-25 06:06:12,939 DEV : loss 2.827807903289795 - f1-score (micro avg) 0.2556
67
+ 2022-08-25 06:06:12,940 BAD EPOCHS (no improvement): 1
68
+ 2022-08-25 06:06:12,941 ----------------------------------------------------------------------------------------------------
69
+ 2022-08-25 06:06:13,054 epoch 3 - iter 1/4 - loss 2.59037260 - samples/sec: 284.59 - lr: 0.100000
70
+ 2022-08-25 06:06:13,081 epoch 3 - iter 2/4 - loss 2.60334442 - samples/sec: 1216.48 - lr: 0.100000
71
+ 2022-08-25 06:06:13,108 epoch 3 - iter 3/4 - loss 2.57638893 - samples/sec: 1264.18 - lr: 0.100000
72
+ 2022-08-25 06:06:13,126 epoch 3 - iter 4/4 - loss 2.53864971 - samples/sec: 1827.04 - lr: 0.100000
73
+ 2022-08-25 06:06:13,126 ----------------------------------------------------------------------------------------------------
74
+ 2022-08-25 06:06:13,127 EPOCH 3 done: loss 2.5386 - lr 0.100000
75
+ 2022-08-25 06:06:13,167 Evaluating as a multi-label problem: False
76
+ 2022-08-25 06:06:13,173 DEV : loss 2.7169787883758545 - f1-score (micro avg) 0.3168
77
+ 2022-08-25 06:06:13,174 BAD EPOCHS (no improvement): 0
78
+ 2022-08-25 06:06:13,175 saving best model
79
+ 2022-08-25 06:06:14,714 ----------------------------------------------------------------------------------------------------
80
+ 2022-08-25 06:06:14,743 epoch 4 - iter 1/4 - loss 2.71329416 - samples/sec: 1140.13 - lr: 0.100000
81
+ 2022-08-25 06:06:14,769 epoch 4 - iter 2/4 - loss 2.38254593 - samples/sec: 1254.42 - lr: 0.100000
82
+ 2022-08-25 06:06:14,803 epoch 4 - iter 3/4 - loss 2.37460295 - samples/sec: 958.15 - lr: 0.100000
83
+ 2022-08-25 06:06:14,825 epoch 4 - iter 4/4 - loss 2.37499146 - samples/sec: 1527.35 - lr: 0.100000
84
+ 2022-08-25 06:06:14,826 ----------------------------------------------------------------------------------------------------
85
+ 2022-08-25 06:06:14,826 EPOCH 4 done: loss 2.3750 - lr 0.100000
86
+ 2022-08-25 06:06:14,864 Evaluating as a multi-label problem: False
87
+ 2022-08-25 06:06:14,870 DEV : loss 2.749277114868164 - f1-score (micro avg) 0.0
88
+ 2022-08-25 06:06:14,871 BAD EPOCHS (no improvement): 1
89
+ 2022-08-25 06:06:14,872 ----------------------------------------------------------------------------------------------------
90
+ 2022-08-25 06:06:14,901 epoch 5 - iter 1/4 - loss 2.32981579 - samples/sec: 1152.14 - lr: 0.100000
91
+ 2022-08-25 06:06:14,928 epoch 5 - iter 2/4 - loss 2.20069920 - samples/sec: 1219.29 - lr: 0.100000
92
+ 2022-08-25 06:06:14,965 epoch 5 - iter 3/4 - loss 2.27754946 - samples/sec: 869.53 - lr: 0.100000
93
+ 2022-08-25 06:06:14,986 epoch 5 - iter 4/4 - loss 2.25410557 - samples/sec: 1547.91 - lr: 0.100000
94
+ 2022-08-25 06:06:14,987 ----------------------------------------------------------------------------------------------------
95
+ 2022-08-25 06:06:14,987 EPOCH 5 done: loss 2.2541 - lr 0.100000
96
+ 2022-08-25 06:06:15,029 Evaluating as a multi-label problem: False
97
+ 2022-08-25 06:06:15,047 DEV : loss 2.7171242237091064 - f1-score (micro avg) 0.2903
98
+ 2022-08-25 06:06:15,049 BAD EPOCHS (no improvement): 2
99
+ 2022-08-25 06:06:15,050 ----------------------------------------------------------------------------------------------------
100
+ 2022-08-25 06:06:15,089 epoch 6 - iter 1/4 - loss 2.18811890 - samples/sec: 858.42 - lr: 0.100000
101
+ 2022-08-25 06:06:15,124 epoch 6 - iter 2/4 - loss 2.14233002 - samples/sec: 929.04 - lr: 0.100000
102
+ 2022-08-25 06:06:15,156 epoch 6 - iter 3/4 - loss 2.20570734 - samples/sec: 1025.55 - lr: 0.100000
103
+ 2022-08-25 06:06:15,173 epoch 6 - iter 4/4 - loss 2.18303889 - samples/sec: 1924.88 - lr: 0.100000
104
+ 2022-08-25 06:06:15,174 ----------------------------------------------------------------------------------------------------
105
+ 2022-08-25 06:06:15,174 EPOCH 6 done: loss 2.1830 - lr 0.100000
106
+ 2022-08-25 06:06:15,210 Evaluating as a multi-label problem: False
107
+ 2022-08-25 06:06:15,217 DEV : loss 2.6913464069366455 - f1-score (micro avg) 0.2314
108
+ 2022-08-25 06:06:15,218 BAD EPOCHS (no improvement): 3
109
+ 2022-08-25 06:06:15,220 ----------------------------------------------------------------------------------------------------
110
+ 2022-08-25 06:06:15,253 epoch 7 - iter 1/4 - loss 2.14348509 - samples/sec: 988.78 - lr: 0.100000
111
+ 2022-08-25 06:06:15,281 epoch 7 - iter 2/4 - loss 2.00850553 - samples/sec: 1178.96 - lr: 0.100000
112
+ 2022-08-25 06:06:15,308 epoch 7 - iter 3/4 - loss 2.06308898 - samples/sec: 1208.59 - lr: 0.100000
113
+ 2022-08-25 06:06:15,327 epoch 7 - iter 4/4 - loss 2.06392736 - samples/sec: 1723.48 - lr: 0.100000
114
+ 2022-08-25 06:06:15,327 ----------------------------------------------------------------------------------------------------
115
+ 2022-08-25 06:06:15,328 EPOCH 7 done: loss 2.0639 - lr 0.100000
116
+ 2022-08-25 06:06:15,368 Evaluating as a multi-label problem: False
117
+ 2022-08-25 06:06:15,376 DEV : loss 2.2334768772125244 - f1-score (micro avg) 0.5
118
+ 2022-08-25 06:06:15,377 BAD EPOCHS (no improvement): 0
119
+ 2022-08-25 06:06:15,381 saving best model
120
+ 2022-08-25 06:06:16,903 ----------------------------------------------------------------------------------------------------
121
+ 2022-08-25 06:06:16,938 epoch 8 - iter 1/4 - loss 1.97582934 - samples/sec: 924.66 - lr: 0.100000
122
+ 2022-08-25 06:06:16,971 epoch 8 - iter 2/4 - loss 1.89160690 - samples/sec: 1003.74 - lr: 0.100000
123
+ 2022-08-25 06:06:17,003 epoch 8 - iter 3/4 - loss 1.92096321 - samples/sec: 1015.15 - lr: 0.100000
124
+ 2022-08-25 06:06:17,024 epoch 8 - iter 4/4 - loss 1.90466607 - samples/sec: 1613.37 - lr: 0.100000
125
+ 2022-08-25 06:06:17,024 ----------------------------------------------------------------------------------------------------
126
+ 2022-08-25 06:06:17,025 EPOCH 8 done: loss 1.9047 - lr 0.100000
127
+ 2022-08-25 06:06:17,076 Evaluating as a multi-label problem: False
128
+ 2022-08-25 06:06:17,086 DEV : loss 2.3985493183135986 - f1-score (micro avg) 0.3448
129
+ 2022-08-25 06:06:17,087 BAD EPOCHS (no improvement): 1
130
+ 2022-08-25 06:06:17,090 ----------------------------------------------------------------------------------------------------
131
+ 2022-08-25 06:06:17,130 epoch 9 - iter 1/4 - loss 1.87079010 - samples/sec: 826.37 - lr: 0.100000
132
+ 2022-08-25 06:06:17,167 epoch 9 - iter 2/4 - loss 1.85216901 - samples/sec: 864.81 - lr: 0.100000
133
+ 2022-08-25 06:06:17,203 epoch 9 - iter 3/4 - loss 1.90496054 - samples/sec: 924.19 - lr: 0.100000
134
+ 2022-08-25 06:06:17,225 epoch 9 - iter 4/4 - loss 1.89038416 - samples/sec: 1452.27 - lr: 0.100000
135
+ 2022-08-25 06:06:17,226 ----------------------------------------------------------------------------------------------------
136
+ 2022-08-25 06:06:17,226 EPOCH 9 done: loss 1.8904 - lr 0.100000
137
+ 2022-08-25 06:06:17,276 Evaluating as a multi-label problem: False
138
+ 2022-08-25 06:06:17,284 DEV : loss 2.121741533279419 - f1-score (micro avg) 0.4627
139
+ 2022-08-25 06:06:17,285 BAD EPOCHS (no improvement): 2
140
+ 2022-08-25 06:06:17,288 ----------------------------------------------------------------------------------------------------
141
+ 2022-08-25 06:06:17,323 epoch 10 - iter 1/4 - loss 1.80540347 - samples/sec: 930.59 - lr: 0.100000
142
+ 2022-08-25 06:06:17,354 epoch 10 - iter 2/4 - loss 1.72573052 - samples/sec: 1054.27 - lr: 0.100000
143
+ 2022-08-25 06:06:17,386 epoch 10 - iter 3/4 - loss 1.77370743 - samples/sec: 1004.27 - lr: 0.100000
144
+ 2022-08-25 06:06:17,407 epoch 10 - iter 4/4 - loss 1.75549973 - samples/sec: 1615.25 - lr: 0.100000
145
+ 2022-08-25 06:06:17,407 ----------------------------------------------------------------------------------------------------
146
+ 2022-08-25 06:06:17,408 EPOCH 10 done: loss 1.7555 - lr 0.100000
147
+ 2022-08-25 06:06:17,453 Evaluating as a multi-label problem: False
148
+ 2022-08-25 06:06:17,461 DEV : loss 2.2982170581817627 - f1-score (micro avg) 0.3333
149
+ 2022-08-25 06:06:17,462 BAD EPOCHS (no improvement): 3
150
+ 2022-08-25 06:06:19,099 ----------------------------------------------------------------------------------------------------
151
+ 2022-08-25 06:06:19,100 loading file db/kgc_models/test-kgc-sm-flair/trained_model/best-model.pt
152
+ 2022-08-25 06:06:27,749 SequenceTagger predicts: Dictionary with 47 tags: O, S-Item, B-Item, E-Item, I-Item, S-Activity, B-Activity, E-Activity, I-Activity, S-Observation, B-Observation, E-Observation, I-Observation, S-Location, B-Location, E-Location, I-Location, S-Consumable, B-Consumable, E-Consumable, I-Consumable, S-Agent, B-Agent, E-Agent, I-Agent, S-Attribute, B-Attribute, E-Attribute, I-Attribute, S-Specifier, B-Specifier, E-Specifier, I-Specifier, S-Cardinality, B-Cardinality, E-Cardinality, I-Cardinality, S-Event, B-Event, E-Event, I-Event, S-Time, B-Time, E-Time, I-Time, <START>, <STOP>
153
+ 2022-08-25 06:06:28,150 Evaluating as a multi-label problem: False
154
+ 2022-08-25 06:06:28,159 0.3281 0.3333 0.3307 0.2386
155
+ 2022-08-25 06:06:28,159
156
+ Results:
157
+ - F-score (micro) 0.3307
158
+ - F-score (macro) 0.125
159
+ - Accuracy 0.2386
160
+
161
+ By class:
162
+ precision recall f1-score support
163
+
164
+ Item 0.2679 0.6250 0.3750 24
165
+ Activity 0.8571 0.6667 0.7500 9
166
+ Observation 0.0000 0.0000 0.0000 13
167
+ Location 0.0000 0.0000 0.0000 5
168
+ Consumable 0.0000 0.0000 0.0000 4
169
+ Agent 0.0000 0.0000 0.0000 3
170
+ Cardinality 0.0000 0.0000 0.0000 2
171
+ Time 0.0000 0.0000 0.0000 2
172
+ Attribute 0.0000 0.0000 0.0000 1
173
+
174
+ micro avg 0.3281 0.3333 0.3307 63
175
+ macro avg 0.1250 0.1435 0.1250 63
176
+ weighted avg 0.2245 0.3333 0.2500 63
177
+
178
+ 2022-08-25 06:06:28,160 ----------------------------------------------------------------------------------------------------