# How to show the learning rate during training
**URL:**
**Author:**  [@milyiyo](/static-proxy?url=https%3A%2F%2Fdiscuss.huggingface.co%2Fu%2Fmilyiyo)%5C
**Post date:** [January 19, 2022, 11:44pm UTC](/static-proxy?url=https%3A%2F%2Fdiscuss.huggingface.co%2Ft%2Fhow-to-show-the-learning-rate-during-training%2F13914%2F1 "2022-01-19T23:44:39Z")
Hi everyone 🙂
I would like to know if it is possible to include the `learning rate` value as part of the information presented during the training.
The columns `Accuracy`, `F1`, `Precision` and `Recall` were added after setting a custom `compute_metrics` function. And I would like to have the `Learning Rate` as well.
Is it possible to add it there?
Thanks in advance 🙂
---
**Author:**  [@Amalq](/static-proxy?url=https%3A%2F%2Fdiscuss.huggingface.co%2Fu%2FAmalq)%5C
**Post date:** [January 22, 2022, 4:29pm UTC](/static-proxy?url=https%3A%2F%2Fdiscuss.huggingface.co%2Ft%2Fhow-to-show-the-learning-rate-during-training%2F13914%2F2 "2022-01-22T16:29:56Z")
Hi Milyiyo,
I am curious how did you add accuracy, F1 Precision and Recall?
---
**Author:**  [@milyiyo](/static-proxy?url=https%3A%2F%2Fdiscuss.huggingface.co%2Fu%2Fmilyiyo)%5C
**Post date:** [January 22, 2022, 7:36pm UTC](/static-proxy?url=https%3A%2F%2Fdiscuss.huggingface.co%2Ft%2Fhow-to-show-the-learning-rate-during-training%2F13914%2F3 "2022-01-22T19:36:36Z")
Hi @Amalq,
I used this:
```python
from sklearn.metrics import precision_recall_fscore_support, accuracy_score
def compute_metrics(pred):
labels = pred.label_ids
preds = pred.predictions.argmax(-1)
precision, recall, f1, _ = precision_recall_fscore_support(labels, preds, average='weighted')
acc = accuracy_score(labels, preds)
return {
'accuracy': acc,
'f1': f1,
'precision': precision,
'recall': recall
}
```
That function was provided to the Trainer like this:
```python
from transformers import Trainer
trainer = Trainer(model=model,
args=training_args,
compute_metrics=compute_metrics,
tokenizer=tokenizer,
eval_dataset=dataset['validation'],
train_dataset=dataset['train'],
data_collator=data_collator)
```
I hope this can help 🙂
---
**Author:**  [@Amalq](/static-proxy?url=https%3A%2F%2Fdiscuss.huggingface.co%2Fu%2FAmalq)%5C
**Post date:** [January 22, 2022, 7:38pm UTC](/static-proxy?url=https%3A%2F%2Fdiscuss.huggingface.co%2Ft%2Fhow-to-show-the-learning-rate-during-training%2F13914%2F4 "2022-01-22T19:38:25Z")
Thank you so much @milyiyo , that is really helpful.
---
**Author:**  [@marshmellow77](/static-proxy?url=https%3A%2F%2Fdiscuss.huggingface.co%2Fu%2Fmarshmellow77)%5C
**Post date:** [January 23, 2022, 9:04am UTC](/static-proxy?url=https%3A%2F%2Fdiscuss.huggingface.co%2Ft%2Fhow-to-show-the-learning-rate-during-training%2F13914%2F5 "2022-01-23T09:04:56Z")
Hi Alberto, yes it is possible to include `learning rate` in the evaluation logs!
Fortunately, the `log()` method of the Trainer class is one of the methods that you can “subclass” to inject custom behaviour: [Trainer](https://huggingface.co/docs/transformers/main_classes/trainer#trainer)
So, all you have to do is create your own Trainer subclass and override the `log()` method like so:
```auto
class MyTrainer(Trainer):
def log(self, logs: Dict[str, float]) -> None:
logs["learning_rate"] = self._get_learning_rate()
super().log(logs)
trainer = MyTrainer(...)
trainer.train()
```
You should now see the learning rate in the eval logs.
Hope that helps, let me know if any questions.
Cheers
Heiko
---
**Author:**  [@milyiyo](/static-proxy?url=https%3A%2F%2Fdiscuss.huggingface.co%2Fu%2Fmilyiyo)%5C
**Post date:** [January 23, 2022, 5:01pm UTC](/static-proxy?url=https%3A%2F%2Fdiscuss.huggingface.co%2Ft%2Fhow-to-show-the-learning-rate-during-training%2F13914%2F6 "2022-01-23T17:01:55Z")
Thanks @marshmellow77 🙂
That solved my problem, look the new column:
Thanks a lot for your help 🤗
---
**Author:**  [@Vighnesh](/static-proxy?url=https%3A%2F%2Fdiscuss.huggingface.co%2Fu%2FVighnesh)%5C
**Post date:** [May 17, 2022, 3:15pm UTC](/static-proxy?url=https%3A%2F%2Fdiscuss.huggingface.co%2Ft%2Fhow-to-show-the-learning-rate-during-training%2F13914%2F7 "2022-05-17T15:15:20Z")
> [@marshmellow77](#):
>
> ```auto
> def log(self, logs: Dict[str, float]) -> None:
> logs["learning_rate"] = self._get_learning_rate()
> super().log(logs)
>
> ```
Hi @marshmellow77 ,
I am trying to use the same logic to log the learning rates at each epoch. But, I am getting the following error:
AttributeError: ‘LMCL\_Trainer’ object has no attribute ‘\_get\_learning\_rate’
LMCL\_Trainer is the name of the subclass extended from Trainer. Providing the class below →
```auto
class LMCL_Trainer(Trainer):
def __init__ (self, *args, **kwargs):
super(). __init__ (*args, **kwargs)
def log(self, logs: Dict[str, float]) -> None:
logs["learning_rate"] = self._get_learning_rate()
super().log(logs)
```
I have the transformers version 3.3.0 installed.
---
**Author:**  [@marshmellow77](/static-proxy?url=https%3A%2F%2Fdiscuss.huggingface.co%2Fu%2Fmarshmellow77)%5C
**Post date:** [May 18, 2022, 10:37am UTC](/static-proxy?url=https%3A%2F%2Fdiscuss.huggingface.co%2Ft%2Fhow-to-show-the-learning-rate-during-training%2F13914%2F8 "2022-05-18T10:37:51Z")
Hi @Vighnesh , the transformers library is currently at version 4.19. Lots has changed since v3.3, I recommend upgrading to the latest version to see if this solution works for you.
Cheers
Heiko
---
**Author:**  [@Vighnesh](/static-proxy?url=https%3A%2F%2Fdiscuss.huggingface.co%2Fu%2FVighnesh)%5C
**Post date:** [May 19, 2022, 6:07am UTC](/static-proxy?url=https%3A%2F%2Fdiscuss.huggingface.co%2Ft%2Fhow-to-show-the-learning-rate-during-training%2F13914%2F9 "2022-05-19T06:07:04Z")
Thanks a lot. Yes the latest version has a lot of changes/improvements.
The solution worked with transformers version 4.19
---
**Author:**  [@matanle](/static-proxy?url=https%3A%2F%2Fdiscuss.huggingface.co%2Fu%2Fmatanle)%5C
**Post date:** [September 26, 2023, 1:44pm UTC](/static-proxy?url=https%3A%2F%2Fdiscuss.huggingface.co%2Ft%2Fhow-to-show-the-learning-rate-during-training%2F13914%2F10 "2023-09-26T13:44:10Z")
Hi 🙂
I’ve tried the suggested solution with transformers version: 4.31.0, but for some reason the learning rate is not plotted as part of the evaluation logs.
Any ideas why? (I do see that it is saved in the log history)
---
**Author:**  [@artyomboyko](/static-proxy?url=https%3A%2F%2Fdiscuss.huggingface.co%2Fu%2Fartyomboyko)%5C
**Post date:** [November 14, 2023, 5:21pm UTC](/static-proxy?url=https%3A%2F%2Fdiscuss.huggingface.co%2Ft%2Fhow-to-show-the-learning-rate-during-training%2F13914%2F11 "2023-11-14T17:21:48Z")
@marshmellow77 Hello. I want to use your solution to infer the learning rate. I am currently using Python 3.10. I get the following error:
```consloe
---------------------------------------------------------------------------
NameError Traceback (most recent call last)
Cell In[36], line 1
----> 1 class MyTrainer(Trainer):
2 def log(self, logs: Dict[str, float]) -> None:
3 logs["learning_rate"] = self._get_learning_rate()
Cell In[36], line 2, in MyTrainer()
1 class MyTrainer(Trainer):
----> 2 def log(self, logs: Dict[str, float]) -> None:
3 logs["learning_rate"] = self._get_learning_rate()
4 print(logs["learning_rate"])
NameError: name 'Dict' is not defined
```
Which specific module do you need to import it from?
---
**Author:**  [@milyiyo](/static-proxy?url=https%3A%2F%2Fdiscuss.huggingface.co%2Fu%2Fmilyiyo)%5C
**Post date:** [January 2, 2024, 1:36pm UTC](/static-proxy?url=https%3A%2F%2Fdiscuss.huggingface.co%2Ft%2Fhow-to-show-the-learning-rate-during-training%2F13914%2F12 "2024-01-02T13:36:34Z")
Hi @artyomboyko,
Did you solve the error?
Maybe was missing something like: `from typing import Dict`
---
**Author:**  [@artyomboyko](/static-proxy?url=https%3A%2F%2Fdiscuss.huggingface.co%2Fu%2Fartyomboyko)%5C
**Post date:** [January 2, 2024, 3:33pm UTC](/static-proxy?url=https%3A%2F%2Fdiscuss.huggingface.co%2Ft%2Fhow-to-show-the-learning-rate-during-training%2F13914%2F13 "2024-01-02T15:33:19Z")
@milyiyo Hello and Happy New year! Yes, you can find solution [here](https://github.com/huggingface/transformers/issues/27631).