hm, but why it happens if my new model has different hyperparameters and is separate experiment with separate name? how to avoid this behavior?

I reinstalled clearml and it works

1.10.3

Yes, I can see it in the blob and it is not empty:

yes, i can see the hyperparams are reflecting correctly, but I mean it shouldn't start from previous model checkpoint if it is different run with different hyperparams

i see, thanks for the explanation, so basically it is initializing the weights to the weights which are saved instead of to the random ones, if i understand correctly? So in this case to prevent this behavior and start from random weights I should just change the name of output file with weights, right?

and previously I was able to retrieve it, I think the trouble started after I finalized dataset.

I also checked the account_name and account_key from clearml.conf file but it seems fine.

But I don't specify pywin in my requirements, so I can't edit its version and I think it is not just a matter of version, but of installing it at all - I've read that you shouldn't install this library on non-Windows machines at all

Yes, but it is not a matter of version - no version of this package can be installed on Linux, this module is available only on Windows.
When I try to install all the packages from my requirements.txt manually on this Linux machine it seems fine, so I am not sure why pywin32 is added as requirement when executing the clearml task.

Thank you John for confirmation. I am getting clearml.storage.helper.StorageError: Can't find azure configuration for azure://...
I did put the storage account name and key in the clearml.conf file as indicated in the documentation and the uri seems to be correct, do you have any idea what I might be missing?

just noticed that even though I put the correct credentials in place, I didn't notice that the whole segment is commented by default :face_palm: it works now, thanks 😁

I see, thanks for explaining

From my end I can confirm that if I use tqdm to print progress of training epochs (like for epoch in tqdm(range(num_epochs)): <training code> ) then I get each tqdm update printed as well as seen in attached image. @<1523702932069945344:profile|CheerfulGorilla72> maybe just reduce the number of updates to be less frequent - I think tqdm parameter miniters allows you to do that.

Still around 8min

ah didn't know the API limit is reseting each month, then it shouldn't be that much of a concern, makes sense. Regarding the metrics and metadata - will they also be counted to the Free Plan 1GB limit if I run my own ClearML self-hosted server?

ok, thanks

Thanks!

It was just mentioning that I don't have access rights

from clearml import Dataset
dataset = Dataset.get(
        dataset_project,
        dataset_name,
        dataset_version)

It is just that - the Dataset is stored in Azure Storage Account.

I was checking my internet speed yesterday and it seemed normal, anyway it seems fine now.

Yes, that could be a workaround, thank you

Hi, I am just having YAML file where I keep all the configurations and hyperparams and then for every training the first thing I do is I call a function which initializes the ClearML task and connects the config and returns it so that the rest of the code can use it.

def initialize_task(task_type=Task.TaskTypes.training):
    config = yaml.safe_load(open(r"params.yaml"))
    Task.ignore_requirements("pywin32")
    task = Task.init(project_name='X', task_name=config['TASK_NAME'], 
     ...

Great, thanks!

it has 24 parameters, but two of them are list of features so overall the yaml file has around 70 lines

Nothing else, is just import and this one line of Dataset.get

ok, that makes sense, thanks a lot for the clarifications

ok, thanks

Request payload:

{"tasks":["425335514d444646b1077cb8b738ccf7","3c0db63779e940cc80895af728aac5ab"],"model_events":false,"no_scroll":true,"iters":1}

I came up with this minimal example - it is a bit different but the behavior is also not as expected I think:

import numpy as np
import plotly.express as px
from clearml import Task

task = Task.init(project_name='MyProject', task_name='task1', 
                 task_type=Task.TaskTypes.training)
logger = task.get_logger()
y_pred = np.random.rand(100)
y_test = np.random.rand(100)
fig = px.line({'y_pred': y_pred, 'y_test': y_test})
logger.report_plotly(title=f'Forecast', series=f'Forecas...