Update README.md
Browse files
README.md
CHANGED
|
@@ -17,6 +17,7 @@ widget:
|
|
| 17 |
|
| 18 |
## Training details:
|
| 19 |
- The training script is provided here: https://github.com/bsinghpratap/roberta_train_goEmotion
|
|
|
|
| 20 |
- The model works well on most of the emotions except: 'desire', 'disgust', 'embarrassment', 'excitement', 'fear', 'grief', 'nervousness', 'pride', 'relief', 'remorse', 'surprise']
|
| 21 |
- I'll try to fine-tune the model further and update here if RoBERTa achieves a better performance.
|
| 22 |
- Each text datapoint can have more than 1 label. Most of the training set had 1 label: Counter({1: 36308, 2: 6541, 3: 532, 4: 28, 5: 1}). So currently I just used the first label for each of the datapoint. Not ideal but it does a decent job.
|
| 17 |
|
| 18 |
## Training details:
|
| 19 |
- The training script is provided here: https://github.com/bsinghpratap/roberta_train_goEmotion
|
| 20 |
+
- Please feel free to start an issue in the repo if you have trouble running the model and I would try to respond as soon as possible.
|
| 21 |
- The model works well on most of the emotions except: 'desire', 'disgust', 'embarrassment', 'excitement', 'fear', 'grief', 'nervousness', 'pride', 'relief', 'remorse', 'surprise']
|
| 22 |
- I'll try to fine-tune the model further and update here if RoBERTa achieves a better performance.
|
| 23 |
- Each text datapoint can have more than 1 label. Most of the training set had 1 label: Counter({1: 36308, 2: 6541, 3: 532, 4: 28, 5: 1}). So currently I just used the first label for each of the datapoint. Not ideal but it does a decent job.
|