Constantly run into this error when installing

[COD1995]

Hi, there,
I followed the instructions in installing the Avalanche library but constantly got this error when running the code, I wonder what is going on:

File ~/miniconda3/envs/avalanche-env/lib/python3.9/site-packages/avalanche/training/templates/observation_type/batch_observation.py:16, in BatchObservation.init(self)
15 def init(self):
---> 16 super().init()
17 self.optimized_param_id = None

TypeError: init() missing 2 required positional arguments: 'model' and 'optimizer'

Thanks a lot. I checked multiple times and ensure I passed in the model and optimizer. I installed avalanche and it is working fine on my local PC but not the case in my server.

[vlomonaco]

Hello, have you checked the avalanche versions are the same on your PC and server?
Can you share a minimal piece of code that reproduces the error?

[COD1995]

Both version is 0.4.0

# Test Setting 1: Clean training dataset
# Clean training dataset and clean test dataset
def train_online_replay(scenario, device, input_size, args):
    model = SlimResNet18(100) # model 
    optimizer = SGD(model.parameters(), lr=args.lr)
    criterion = CrossEntropyLoss()

    interactive_logger = InteractiveLogger() # yet to be explored 
    loggers = [interactive_logger]

    training_metrics = []
    evaluation_metrics = [
        accuracy_metrics(epoch=True, stream=True),
        loss_metrics(epoch=True, stream=True),
    ]

    evaluator = EvaluationPlugin(
        *training_metrics,
        *evaluation_metrics,
        loggers=loggers,
    )
    
    storage_policy = ClassBalancedBuffer(args.mem_size, adaptive_size=True)
    plugins = [ReplayPlugin(args.mem_size, storage_policy=storage_policy)]

    cl_strategy = LwF(
        model=model,
        optimizer=optimizer,
        criterion=criterion,
        train_epochs= 100,
        alpha = 1.0,
        temperature=2.0,
        plugins=plugins,
        evaluator=evaluator,
        device=device,
        train_mb_size=args.train_mb_size,
        eval_mb_size=64,
    )

    # For online scenario
    batch_streams = scenario.streams.values()
    
    # my modification
    results = []
    
    for t, experience in enumerate(scenario.train_stream):
        print("Start of experience: ", experience.current_experience)
        print("Current Classes: ", experience.classes_in_this_experience)

        ocl_scenario = OnlineCLScenario(
            original_streams=batch_streams,
            experiences=experience,
            experience_size=args.train_mb_size,
            access_task_boundaries=False,
        )

        # Set this to inform strat
        cl_strategy.classes_in_this_experience = experience.classes_in_this_experience

        cl_strategy.train(
            ocl_scenario.train_stream,
            eval_streams=[],
            num_workers=0,
            drop_last=True,
        )
        
        # print(f"Evaluating on test data of experience 0 using cl_strategy.eval")
        # eval_results = cl_strategy.eval(scenario.test_stream[0])
        # strategy_accuracy = eval_results['Top1_Acc_Stream/eval_phase/test_stream/Task000']
        # print(f"Accuracy using cl_strategy.eval: {strategy_accuracy * 100:.2f}%")
        # results.append(strategy_accuracy)
        
        # Save model weights after training on each experience
        model_save_path = f'clean_model_weights_experience_{t}.pth'
        torch.save(model.state_dict(), model_save_path)
        print(f'Model weights saved to {model_save_path}')
        
    # Plotting the results
    visualize_results(results)
    
    return results

[AlbinSou]

Damn, it's the typing-extensions library error.

You can try

pip install typing-extensions==4.4.0

maybe it will fix it. But I thought that we had corrected it, the correct version of typing extension is in the requirements file ... What command did you run for the installation ?

[COD1995]

I tried both:

pip install avalanche-lib

and

pip install avalanche-lib[all]

[COD1995]

It does not fix the issue:

Requirement already satisfied: typing-extensions==4.4.0 in ./miniconda3/envs/avalanche-env/lib/python3.9/site-packages (4.4.0)

[AlbinSou]

Can you give me the output of conda list|grep typing ? It can happen that this library is installed in multiple ways, through pip and conda, and that the wrong one is used. It just happened to me today.

[AntonioCarta]

@COD1995 you need to start from a clean environment.