progress bars not displaying in vscode Jupyter notepad

[codermrrob]

A few days ago fastprogress stopped working in Jupyter notepads in vscode.

It seems to run without error but does not display anything. I have the same issue in existing environments (venv) and in newly created environments. I have reverted to Jan 2023 version of vscode where I know it was working before, but that showed no change.

Running the following code works for console output, but in Jupyter notebook nothing.

Where should I look to try and resolve this issue? Any guidance will be valuable because I have pretty much exhausted my knowledge of fastprogress and what I can do.

from time import sleep
from fastprogress.fastprogress import master_bar, progress_bar

mb = master_bar(range(10))
progress_bar(range(100), parent=mb)

for i in mb:
    for j in progress_bar(range(100), parent=mb):
        sleep(0.01)

[elisgrahn]

I have the exact same problem. Discovered something was wrong since fastai wasn't displaying any bars. Tqdm for instance, still works in notebooks.

I have also tried to revert to Jan 2023 vscode and also reverted ipywidgets to 7.7.* instead of 8 without any further luck.

[codermrrob]

@ElisGrahn I also reverted ipywidgets, but I can verify both the ipywidgets 7.7.* and 8 work fine; I can import and run and successfully display widgets in notebooks that I have tried. fastprogress is working in console mode, of course, but not in notebooks in vscode.

makes working through the fastai courses problematic with local setup, which is disappointing because I want to utilize my GPU (which I have successfully set up, which is the trickier part). But now I've hit this wall.

[peterbull]

Also having this issue. In the meantime, if anyone like myself is still early in the learning process and just needs the epoch data, there is a quick workaround. Add cbs=CSVLogger() to your learner, and then create a quick function to display it. Example:

def train_res():
    train_data = pd.read_csv('history.csv')
    return train_data.head(n=50)

learn = tabular_learner(dls, layers=[500, 250], n_out=1, y_range=(-11,11), loss_func=F.mse_loss, cbs=CSVLogger())

learn.fit_one_cycle(1, 1e-2)
train_res()

[elisgrahn]

Nice workaround! Just want to share that the solution I've currently settled on is to add a TensorBoardCallback() in order to view the models in Tensorboard. Might be a bit overkill, but I felt it was useful to learn how to use it.

cbs=[TensorBoardCallback(n_preds=8)]

[bosmart]

Same issue here - no progress bars in vscode :(

[bosmart]

Ok, I tried a few different versions of the Jupyter extension and: v2022.11.1003412109 - working, ShowGraphCallback works too v2023.1.2000312134 - working, ShowGraphCallback doesn’t (plot disappears) v2023.1.2010391206 - working, ShowGraphCallback doesn’t (plot disappears) v2023.2.1000592019 - not working, ShowGraphCallback works again v2023.2.1200692131 - not working, ShowGraphCallback works again

[codermrrob]

@bosmart thanks for the effort there, I hadn't been aware it was so easy to revert extension versions in vscode now.

[bosmart]

@codermrrob I've opened the following: https://github.com/microsoft/vscode-jupyter/issues/13163

[aisbergg]

Same issue here. Just started with the course and was wondering why I don't get any output. I played around and found out that with a short sleep between the display.update() calls in the show method of the progress bar it actually outputs something. Might be a race condition somewhere.

Here is my code snippet (I only inserted some time.sleep(), the rest is basically unchanged):

# workaround for displaying progress bar in notebook
import time
from fastai.vision.all import *
from fastprogress.fastprogress import NBMasterBar, NBProgressBar

class NBMasterBarPatched(NBMasterBar):
    def on_iter_begin(self):
        super().on_iter_begin()
        time.sleep(0.01)

    def show(self):
        time.sleep(0.01)
        super().show()

class NBProgressBarPatched(NBProgressBar):
    def on_iter_begin(self):
        super().on_iter_begin()
        time.sleep(0.01)

    def show(self):
        time.sleep(0.01)
        super().show()

class ProgressCallbackPatched(ProgressCallback):
    def before_fit(self):
        if self.create_mbar: self.mbar = NBMasterBarPatched(list(range(self.n_epoch)))
        assert hasattr(self.learn, 'recorder')
        if self.learn.logger != noop:
            self.old_logger,self.learn.logger = self.logger,self._write_stats
            self._write_stats(self.recorder.metric_names)
        else: self.old_logger = noop

    def _launch_pbar(self):
        self.pbar = NBProgressBarPatched(self.dl, parent=getattr(self, 'mbar', None), leave=False)
        self.pbar.update(0)

defaults.callbacks = [TrainEvalCallback, Recorder, ProgressCallbackPatched]

# now do your learning
learn...

[adam-matic]

Another workaround from an older issue, forcing text output for the progress bar:

from fastprogress.fastprogress import master_bar, progress_bar
from fastprogress.fastprogress import force_console_behavior
master_bar, progress_bar = force_console_behavior()

[pounde]

Same issue here. Reverted jupyter extension and it seems to be working. I'll add a comment to the vs-code jupyter issue as well.

EDIT: Should have looked at the jupyter issue first. Looks like it was merged two days ago. Guessing we'll see it in the next release.