AttributeError: 'NoneType' object has no attribute 'log_metric'

[Zessay]

I have no idea about the following error, looking forward to someone's help, thank you! ~~

Traceback (most recent call last):
  File "disfluency_detection/pipelines/sentseg_train.py", line 427, in <module>
    main()
  File "disfluency_detection/pipelines/sentseg_train.py", line 394, in main
    metrics = trainer.evaluate()
  File "/home/xxx/anaconda3/envs/correction/lib/python3.7/site-packages/transformers/trainer.py", line 2044, in evaluate
    self.log(output.metrics)
  File "/home/xxx/anaconda3/envs/correction/lib/python3.7/site-packages/transformers/trainer.py", line 1709, in log
    self.control = self.callback_handler.on_log(self.args, self.state, self.control, logs)
  File "/home/xxx/anaconda3/envs/correction/lib/python3.7/site-packages/transformers/trainer_callback.py", line 371, in on_log
    return self.call_event("on_log", args, state, control, logs=logs)
  File "/home/xxx/anaconda3/envs/correction/lib/python3.7/site-packages/transformers/trainer_callback.py", line 388, in call_event
    **kwargs,
  File "/home/xxx/anaconda3/envs/correction/lib/python3.7/site-packages/transformers/integrations.py", line 354, in on_log
    self.tb_writer.add_scalar(k, v, state.global_step)
  File "/home/xxx/anaconda3/envs/correction/lib/python3.7/site-packages/tensorboardX/writer.py", line 453, in add_scalar
    self.comet_logger.log_metric(tag, display_name, scalar_value, global_step)
AttributeError: 'NoneType' object has no attribute 'log_metric'

[Zessay]

I have found the solution to this problem in one transformers issue

[lanpa]

Thanks for the reporting, seems that it's caused by the new introduced feature. Let me reopen this issue for tracking.

[syomantak]

The error still persists! The commit didn't fix it.

[cdhx]

same problem

[syomantak]

@cdhx I found the error for my code - as I was using Jupyter notebook, I didn't realize I closed the writer and then tried to write again in another cell.

[lanpa]

@cdhx @CyanideBoy Hi, which version are you using?

[grafaraway]

I used with summary_writer:summary_writer.add_graph( ) before .add_scalar( )
After deleting with summary_writer: , my code could get correct output.

[mwck46]

I encountered a similar problem: 'NoneType' object has no attribute 'log_metrics'.
The code works fine after deleting with writer: like grafaraway suggested.
My package version:
pytorch 1.8.2 py3.8_cpu_0 [cpuonly] pytorch-lts
tensorboardx 2.4 pyhd8ed1ab_0 conda-forge

[lanpa]

@mwck46 Can you provide a code snippet to reproduce the error with with...? Thank you.

[mwck46]

@lanpa hope this help

import torch
import torch.nn as nn
import torch.utils.data as dset
from torchvision import datasets, transforms
from tensorboardX import SummaryWriter  

dataset_path = "./data"
log_path = "./log"
device = "cpu"

class ANNModel(nn.Module):
    def __init__(self):
        super(ANNModel, self).__init__()
        self.main = nn.Sequential(
            nn.Linear(in_features=28*28, out_features=128),
            nn.ReLU(),
            nn.Linear(in_features=128, out_features=10),
            nn.LogSoftmax(dim=1))
    def forward(self, input):
        return self.main(input)

def load_data(batch_size, transform):
    trainset = datasets.MNIST(root=dataset_path, download=True, train=True, transform=transform)
    testset = datasets.MNIST(root=dataset_path, download=True, train=False, transform=transform)
    trainloader = dset.DataLoader(trainset, batch_size=batch_size, shuffle=True)
    testloader = dset.DataLoader(testset, batch_size=batch_size, shuffle=False)
    return trainloader, testloader

def main():
    writer = SummaryWriter(log_path)
    
    transform = transforms.Compose([transforms.ToTensor(),])
    trainloader, testloader = load_data(batch_size = 64, transform=transform)
    model = ANNModel().to(device)
    train_data_sample, _ = iter(trainloader).next()
    
    with writer:
        train_data_sample = train_data_sample.view(train_data_sample.shape[0], -1)
        writer.add_graph(model, train_data_sample.to(device))  # model graph, with input

    epochs = 3
    for epoch in range(epochs):
        epoch_loss = 0.1
        
        writer.add_scalars("Loss/Epoch", {"train": epoch_loss}, epoch)
        print("epoch[%02d/%02d] | training loss: %.4f" % (epoch + 1, epochs, epoch_loss))

if __name__ == "__main__":
    main()

My conda env env.txt

[lanpa]

@mwck46 According to your env.txt, maybe you are using the previous tensorboardX (v2.4) from conda-forge
Will pip install 'git+https://github.com/lanpa/tensorboardX' resolve your problem?

[DrraBL]

Hi everyone
I got the same error with tensorboard 2.4. someone could help me to solve it, please?
I appreciate your time

[lanpa]

@DrraBL Can you try whether pip install git+https://github.com/lanpa/tensorboardX helps?