YagiASAFAS
commited on
Commit
•
f30fa89
1
Parent(s):
9d152bd
update model card README.md
Browse files
README.md
CHANGED
@@ -14,17 +14,10 @@ should probably proofread and complete it, then remove this comment. -->
|
|
14 |
|
15 |
# polibert-malaysia-ver4
|
16 |
|
17 |
-
This model is
|
18 |
-
What is new is that this model used a new dataset which not only used tnwei/ms-newspapers dataset but also almost 10k of instagram posts regarding several topics about Malaysia.
|
19 |
-
By doing so, this model captures not only formal sentences such as News, but also captures informal sentences such as personal posts.
|
20 |
-
As a tradeoff of ver3, the accuracy was quite lower compared to the previous one(ver2).
|
21 |
-
This time we extracted data which have stronger characteristics regarding text features from the dataset used in ver3.
|
22 |
-
By doing so, now this model captures not only formal sentences such as News, but also captures informal sentences such as personal posts.
|
23 |
-
And also the accuracy is higher than the previous one(ver).
|
24 |
-
However, several data were deleted from the dataset. Thus the devirsity has decreased. This is the tradeoff for this time.
|
25 |
It achieves the following results on the evaluation set:
|
26 |
-
- Loss: 0.
|
27 |
-
- Accuracy: 0.
|
28 |
|
29 |
## Model description
|
30 |
|
@@ -45,38 +38,25 @@ More information needed
|
|
45 |
The following hyperparameters were used during training:
|
46 |
- learning_rate: 3e-05
|
47 |
- train_batch_size: 8
|
48 |
-
- eval_batch_size:
|
49 |
- seed: 42
|
50 |
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
|
51 |
- lr_scheduler_type: linear
|
52 |
- num_epochs: 8
|
53 |
- mixed_precision_training: Native AMP
|
54 |
|
55 |
-
### Label Mappings
|
56 |
-
- 0: Economic Concerns
|
57 |
-
- 1: Racial discrimination or polarization
|
58 |
-
- 2: Leadership weaknesses
|
59 |
-
- 3: Development and infrastructure gaps
|
60 |
-
- 4: Corruption
|
61 |
-
- 5: Political instablility
|
62 |
-
- 6: Socials and Public safety
|
63 |
-
- 7: Administration
|
64 |
-
- 8: Education
|
65 |
-
- 9: Religion issues
|
66 |
-
- 10: Environmental
|
67 |
-
|
68 |
### Training results
|
69 |
|
70 |
-
| Training Loss | Epoch | Step
|
71 |
-
|
72 |
-
|
|
73 |
-
| 0.
|
74 |
-
| 0.
|
75 |
-
| 0.
|
76 |
-
| 0.
|
77 |
-
| 0.
|
78 |
-
| 0.
|
79 |
-
| 0.
|
80 |
|
81 |
|
82 |
### Framework versions
|
|
|
14 |
|
15 |
# polibert-malaysia-ver4
|
16 |
|
17 |
+
This model is a fine-tuned version of [bert-base-uncased](https://huggingface.co/bert-base-uncased) on an unknown dataset.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
18 |
It achieves the following results on the evaluation set:
|
19 |
+
- Loss: 0.2641
|
20 |
+
- Accuracy: 0.9371
|
21 |
|
22 |
## Model description
|
23 |
|
|
|
38 |
The following hyperparameters were used during training:
|
39 |
- learning_rate: 3e-05
|
40 |
- train_batch_size: 8
|
41 |
+
- eval_batch_size: 8
|
42 |
- seed: 42
|
43 |
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
|
44 |
- lr_scheduler_type: linear
|
45 |
- num_epochs: 8
|
46 |
- mixed_precision_training: Native AMP
|
47 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
48 |
### Training results
|
49 |
|
50 |
+
| Training Loss | Epoch | Step | Validation Loss | Accuracy |
|
51 |
+
|:-------------:|:-----:|:----:|:---------------:|:--------:|
|
52 |
+
| 1.032 | 1.0 | 938 | 0.3282 | 0.9227 |
|
53 |
+
| 0.2967 | 2.0 | 1876 | 0.2641 | 0.9371 |
|
54 |
+
| 0.1987 | 3.0 | 2814 | 0.2902 | 0.9403 |
|
55 |
+
| 0.163 | 4.0 | 3752 | 0.2995 | 0.9451 |
|
56 |
+
| 0.1315 | 5.0 | 4690 | 0.2922 | 0.9445 |
|
57 |
+
| 0.0864 | 6.0 | 5628 | 0.2760 | 0.9504 |
|
58 |
+
| 0.0861 | 7.0 | 6566 | 0.2836 | 0.9493 |
|
59 |
+
| 0.0686 | 8.0 | 7504 | 0.2933 | 0.9509 |
|
60 |
|
61 |
|
62 |
### Framework versions
|