vibhorag101
commited on
Commit
•
28c4143
1
Parent(s):
02d93ec
Update README.md
Browse files
README.md
CHANGED
@@ -10,15 +10,37 @@ metrics:
|
|
10 |
- f1
|
11 |
model-index:
|
12 |
- name: roberta-base-suicide-prediction-phr-v2
|
13 |
-
results:
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
14 |
---
|
15 |
|
|
|
16 |
<!-- This model card has been generated automatically according to the information the Trainer had access to. You
|
17 |
should probably proofread and complete it, then remove this comment. -->
|
18 |
|
19 |
# vibhorag101/roberta-base-suicide-prediction-phr-v2
|
20 |
|
21 |
-
This model is a fine-tuned version of [roberta-base](https://huggingface.co/roberta-base) on
|
22 |
It achieves the following results on the evaluation set:
|
23 |
- Loss: 0.0553
|
24 |
- Accuracy: 0.9869
|
@@ -27,21 +49,24 @@ It achieves the following results on the evaluation set:
|
|
27 |
- F1: 0.9875
|
28 |
|
29 |
## Model description
|
30 |
-
|
31 |
-
More information needed
|
32 |
-
|
33 |
-
## Intended uses & limitations
|
34 |
-
|
35 |
-
More information needed
|
36 |
|
37 |
## Training and evaluation data
|
38 |
-
|
39 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
40 |
|
41 |
## Training procedure
|
|
|
42 |
|
43 |
### Training hyperparameters
|
44 |
-
|
45 |
The following hyperparameters were used during training:
|
46 |
- learning_rate: 2e-05
|
47 |
- train_batch_size: 16
|
@@ -51,6 +76,12 @@ The following hyperparameters were used during training:
|
|
51 |
- lr_scheduler_type: linear
|
52 |
- lr_scheduler_warmup_ratio: 0.06
|
53 |
- num_epochs: 3
|
|
|
|
|
|
|
|
|
|
|
|
|
54 |
|
55 |
### Training results
|
56 |
|
|
|
10 |
- f1
|
11 |
model-index:
|
12 |
- name: roberta-base-suicide-prediction-phr-v2
|
13 |
+
results:
|
14 |
+
- task:
|
15 |
+
type: text-classification
|
16 |
+
name: Suicidal Tendency Prediction in text
|
17 |
+
dataset:
|
18 |
+
type: vibhorag101/phr_suicide_prediction_dataset_clean_light
|
19 |
+
name: Suicide Prediction Dataset
|
20 |
+
split: val
|
21 |
+
metrics:
|
22 |
+
- type: accuracy
|
23 |
+
value: 0.9869
|
24 |
+
- type: f1
|
25 |
+
value: 0.9875
|
26 |
+
- type: recall
|
27 |
+
value: 0.9846
|
28 |
+
- type: precision
|
29 |
+
value: 0.9904
|
30 |
+
datasets:
|
31 |
+
- vibhorag101/phr_suicide_prediction_dataset_clean_light
|
32 |
+
language:
|
33 |
+
- en
|
34 |
+
library_name: transformers
|
35 |
---
|
36 |
|
37 |
+
|
38 |
<!-- This model card has been generated automatically according to the information the Trainer had access to. You
|
39 |
should probably proofread and complete it, then remove this comment. -->
|
40 |
|
41 |
# vibhorag101/roberta-base-suicide-prediction-phr-v2
|
42 |
|
43 |
+
This model is a fine-tuned version of [roberta-base](https://huggingface.co/roberta-base) on [Suicide Prediction Dataset](https://huggingface.co/datasets/vibhorag101/phr_suicide_prediction_dataset_clean_light), sourced from Reddit.
|
44 |
It achieves the following results on the evaluation set:
|
45 |
- Loss: 0.0553
|
46 |
- Accuracy: 0.9869
|
|
|
49 |
- F1: 0.9875
|
50 |
|
51 |
## Model description
|
52 |
+
This model is a finetune of roberta-base to detect suicidal tendencies in a given text.
|
|
|
|
|
|
|
|
|
|
|
53 |
|
54 |
## Training and evaluation data
|
55 |
+
- The dataset is sourced from Reddit and is available on [Kaggle](https://www.kaggle.com/datasets/nikhileswarkomati/suicide-watch).
|
56 |
+
- The dataset contains text with binary labels for suicide or non-suicide.
|
57 |
+
- The dataset was cleaned minimally, as BERT depends on contextually sensitive information, which can worsely effect its performance.
|
58 |
+
- Removed numbers
|
59 |
+
- Removed URLs, Emojis, and accented characters.
|
60 |
+
- Remove any extra white spaces and any extra spaces after a single space.
|
61 |
+
- Removed any consecutive characters repeated more than 3 times.
|
62 |
+
- The rows with more than 512 BERT Tokens were removed, as they exceeded BERT's max token.
|
63 |
+
- The cleaned dataset can be found [here](https://huggingface.co/datasets/vibhorag101/phr_suicide_prediction_dataset_clean_light)
|
64 |
+
- The evaluation set had ~33k samples, while the training set had ~153k samples, i.e., a 70:15:15 (train:test:val) split.
|
65 |
|
66 |
## Training procedure
|
67 |
+
- The model was trained on an RTXA5000 GPU.
|
68 |
|
69 |
### Training hyperparameters
|
|
|
70 |
The following hyperparameters were used during training:
|
71 |
- learning_rate: 2e-05
|
72 |
- train_batch_size: 16
|
|
|
76 |
- lr_scheduler_type: linear
|
77 |
- lr_scheduler_warmup_ratio: 0.06
|
78 |
- num_epochs: 3
|
79 |
+
- eval_steps: 500
|
80 |
+
- save_steps: 500
|
81 |
+
- Early Stopping:
|
82 |
+
- early_stopping_patience: 5
|
83 |
+
- early_stopping_threshold: 0.001
|
84 |
+
- parameter: F1 Score
|
85 |
|
86 |
### Training results
|
87 |
|