search

AlbinSou / ocl_survey

Code for "A Comprehensive Empirical Evaluation on Online Continual Learning" ICCVW 2023 VCL Workshop
https://arxiv.org/abs/2308.10328
MIT License
37 stars 9 forks source link

AssertionError: no class means available for icarl strategy #9

Closed hiteshvaidya closed 10 months ago

hiteshvaidya commented 1 year ago

I set n_experiences=10 for split-cifar10 and task_boundaries=False at line 98 in main.py. I get the following error while running the experiment

(ocl_survey) [hvaidya@forest.usf.edu@GPU6 experiments]$ HYDRA_FULL_ERROR=1 python main.py strategy=icarl experiment=split_cifar10 evaluation=parallel strategy.train_epochs=1 strategy.mem_size=2000                
/data/hvaidya/ocl_survey/experiments/main.py:21: UserWarning:                                                                                                                                                       
The version_base parameter is not specified.                                                                                                                                                                        
Please specify a compatability version level, or None.                                                                                                                                                              
Will assume defaults for version 1.1                                                                                                                                                                                
  @hydra.main(config_path="../config", config_name="config.yaml")                                                                                                                                                   
/home/h/hvaidya/.conda/envs/ocl_survey/lib/python3.10/site-packages/hydra/_internal/hydra.py:119: UserWarning: Future Hydra versions will no longer change working directory at job runtime by default.             
See https://hydra.cc/docs/1.2/upgrades/1.1_to_1.2/changes_to_job_working_dir/ for more information.                                                                                                                 
  ret = run_job(                                                                                                                                                                                                    
Files already downloaded and verified                                                                                                                                                                               
Files already downloaded and verified                                                                                                                                                                               
[4, 1, 7, 5, 3, 9, 0, 8, 6, 2]                                                                                                                                                                                      
SGD (                                                                                                                                                                                                               
Parameter Group 0                                                                                                                                                                                                   
    dampening: 0                                                                                                                                                                                                    
    differentiable: False                                                                                                                                                                                           
    foreach: None                                                                                                                                                                                                   
    lr: 0.1                                                                                                                                                                                                         
    maximize: False                                                                                                                                                                                                 
    momentum: 0.0                                                                                                                                                                                                   
    nesterov: False                                                                                                                                                                                                 
    weight_decay: 0.0                                                                                                                                                                                               
)                                                                                                                                                                                                                   
Using strategy:  OnlineICaRL                                                                                                                                                                                        
With plugins:  [<src.toolkit.parallel_eval.ParallelEvaluationPlugin object at 0x7f41779fee00>, <src.strategies.icarl._ICaRLPlugin object at 0x7f41434727a0>, <avalanche.training.plugins.replay.ReplayPlugin object 
at 0x7f41434731f0>, <src.strategies.icarl.OnlineICaRLLossPlugin object at 0x7f4177bb0f10>, <avalanche.training.plugins.evaluation.EvaluationPlugin object at 0x7f4177bb0b80>, <avalanche.training.templates.base_sgd
.PeriodicEval object at 0x7f4143472890>, <avalanche.training.plugins.clock.Clock object at 0x7f41434728f0>]                                                                                                         
-- >> Start of training phase << --                                                                                                                                                                                 
-- Starting training on experience 0 (Task 0) from train stream --                                                                                                                                                  
  0%|▎                                                                                                                                                                              | 1/475 [00:03<25:52,  3.28s/it]
Error executing job with overrides: ['strategy=icarl', 'experiment=split_cifar10', 'evaluation=parallel', 'strategy.train_epochs=1', 'strategy.mem_size=2000']                                                      
Traceback (most recent call last):                                                                                                                                                                                  
  File "/data/hvaidya/ocl_survey/experiments/main.py", line 129, in <module>                                                                                                                                        
    main()                                                                                                                                                                                                          
  File "/home/h/hvaidya/.conda/envs/ocl_survey/lib/python3.10/site-packages/hydra/main.py", line 94, in decorated_main                                                                                              
    _run_hydra(                                                                                                                                                                                                     
  File "/home/h/hvaidya/.conda/envs/ocl_survey/lib/python3.10/site-packages/hydra/_internal/utils.py", line 394, in _run_hydra                                                                                      
    _run_app(                                                                                                                                                                                                       
  File "/home/h/hvaidya/.conda/envs/ocl_survey/lib/python3.10/site-packages/hydra/_internal/utils.py", line 457, in _run_app                                                                                        
    run_and_report(                                                                                                                                                                                                 
  File "/home/h/hvaidya/.conda/envs/ocl_survey/lib/python3.10/site-packages/hydra/_internal/utils.py", line 223, in run_and_report                                                                                  
    raise ex                                                                                                                                                                                                        
  File "/home/h/hvaidya/.conda/envs/ocl_survey/lib/python3.10/site-packages/hydra/_internal/utils.py", line 220, in run_and_report                                                                                  
    return func()                                                                                                                                                                                                   
  File "/home/h/hvaidya/.conda/envs/ocl_survey/lib/python3.10/site-packages/hydra/_internal/utils.py", line 458, in <lambda>                                                                                        
    lambda: hydra.run(                                                                                                                                                                                              
  File "/home/h/hvaidya/.conda/envs/ocl_survey/lib/python3.10/site-packages/hydra/_internal/hydra.py", line 132, in run                                                                                             
    _ = ret.return_value                                                                                                                                                                                            
  File "/home/h/hvaidya/.conda/envs/ocl_survey/lib/python3.10/site-packages/hydra/core/utils.py", line 260, in return_value                                                                                         
    raise self._return_value                                                                                                                                                                                        
  File "/home/h/hvaidya/.conda/envs/ocl_survey/lib/python3.10/site-packages/hydra/core/utils.py", line 186, in run_job                                                                                              
    ret.return_value = task_function(task_cfg)                                                                                                                                                                      
  File "/data/hvaidya/ocl_survey/experiments/main.py", line 109, in main                                                                                                                                            
    strategy.train(                                                                                                                                                                                                 
  File "/home/h/hvaidya/.conda/envs/ocl_survey/lib/python3.10/site-packages/avalanche/training/templates/base_sgd.py", line 168, in train                                                                           
    super().train(experiences, eval_streams, **kwargs)                                                                                                                                                              
  File "/home/h/hvaidya/.conda/envs/ocl_survey/lib/python3.10/site-packages/avalanche/training/templates/base.py", line 144, in train                                                                               
    self._train_exp(self.experience, eval_streams, **kwargs)                                                                                                         
  File "/home/h/hvaidya/.conda/envs/ocl_survey/lib/python3.10/site-packages/avalanche/training/templates/base_sgd.py", line 294, in _train_exp                                                                      
    self.training_epoch(**kwargs)
  File "/home/h/hvaidya/.conda/envs/ocl_survey/lib/python3.10/site-packages/avalanche/training/templates/update_type/sgd_update.py", line 22, in training_epoch
    self._before_forward(**kwargs)
  File "/home/h/hvaidya/.conda/envs/ocl_survey/lib/python3.10/site-packages/avalanche/training/templates/base_sgd.py", line 473, in _before_forward
    trigger_plugins(self, "before_forward", **kwargs)
  File "/home/h/hvaidya/.conda/envs/ocl_survey/lib/python3.10/site-packages/avalanche/training/utils.py", line 69, in trigger_plugins
    getattr(p, event)(strategy, **kwargs)
  File "/data/hvaidya/ocl_survey/src/strategies/icarl.py", line 52, in before_forward
    self.old_logits = self.old_model(strategy.mb_x)
  File "/home/h/hvaidya/.conda/envs/ocl_survey/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1518, in _wrapped_call_impl
    return self._call_impl(*args, **kwargs)
  File "/home/h/hvaidya/.conda/envs/ocl_survey/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1527, in _call_impl
    return forward_call(*args, **kwargs)
  File "/home/h/hvaidya/.conda/envs/ocl_survey/lib/python3.10/site-packages/avalanche/models/dynamic_modules.py", line 460, in forward
    return self.eval_classifier(x)
  File "/home/h/hvaidya/.conda/envs/ocl_survey/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1518, in _wrapped_call_impl
    return self._call_impl(*args, **kwargs)
  File "/home/h/hvaidya/.conda/envs/ocl_survey/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1527, in _call_impl
    return forward_call(*args, **kwargs)
  File "/home/h/hvaidya/.conda/envs/ocl_survey/lib/python3.10/site-packages/torch/utils/_contextlib.py", line 115, in decorate_context
    return func(*args, **kwargs)
  File "/home/h/hvaidya/.conda/envs/ocl_survey/lib/python3.10/site-packages/avalanche/models/ncm_classifier.py", line 77, in forward
    assert self.class_means_dict != {}, "no class means available."
AssertionError: no class means available.
Exception ignored in atexit callback: <bound method BlockingScheduler.close of <src.toolkit.parallel_eval.BlockingScheduler object at 0x7f4177b5df60>>
Traceback (most recent call last):
  File "/data/hvaidya/ocl_survey/src/toolkit/parallel_eval.py", line 37, in close
    ray.get(self.scheduled_tasks)
  File "/home/h/hvaidya/.conda/envs/ocl_survey/lib/python3.10/site-packages/ray/_private/auto_init_hook.py", line 24, in auto_init_wrapper
    return fn(*args, **kwargs)
  File "/home/h/hvaidya/.conda/envs/ocl_survey/lib/python3.10/site-packages/ray/_private/client_mode_hook.py", line 103, in wrapper
    return func(*args, **kwargs)
  File "/home/h/hvaidya/.conda/envs/ocl_survey/lib/python3.10/site-packages/ray/_private/worker.py", line 2524, in get
    raise value.as_instanceof_cause()
ray.exceptions.RayTaskError(AssertionError): ray::EvaluationActor.eval() (pid=959167, ip=10.10.10.6, actor_id=6004a1dcca0184816d26c34101000000, repr=<src.toolkit.parallel_eval.EvaluationActor object at 0x7f05f22b
6e30>)
  File "/data/hvaidya/ocl_survey/src/toolkit/parallel_eval.py", line 66, in eval
    self.strat.eval(stream, **kwargs)
  File "/home/h/hvaidya/.conda/envs/ocl_survey/lib/python3.10/site-packages/torch/utils/_contextlib.py", line 115, in decorate_context
    return func(*args, **kwargs)
  File "/home/h/hvaidya/.conda/envs/ocl_survey/lib/python3.10/site-packages/avalanche/training/templates/base_sgd.py", line 183, in eval
    super().eval(exp_list, **kwargs)
  File "/home/h/hvaidya/.conda/envs/ocl_survey/lib/python3.10/site-packages/torch/utils/_contextlib.py", line 115, in decorate_context
    return func(*args, **kwargs)
  File "/home/h/hvaidya/.conda/envs/ocl_survey/lib/python3.10/site-packages/avalanche/training/templates/base.py", line 193, in eval
    self._eval_exp(**kwargs)
  File "/home/h/hvaidya/.conda/envs/ocl_survey/lib/python3.10/site-packages/avalanche/training/templates/base_sgd.py", line 187, in _eval_exp
    self.eval_epoch(**kwargs)
  File "/home/h/hvaidya/.conda/envs/ocl_survey/lib/python3.10/site-packages/avalanche/training/templates/base_sgd.py", line 229, in eval_epoch
    self.mb_output = self.forward()
  File "/home/h/hvaidya/.conda/envs/ocl_survey/lib/python3.10/site-packages/avalanche/training/templates/problem_type/supervised_problem.py", line 39, in forward
    return avalanche_forward(self.model, self.mb_x, self.mb_task_id)
  File "/home/h/hvaidya/.conda/envs/ocl_survey/lib/python3.10/site-packages/avalanche/models/utils.py", line 21, in avalanche_forward
    return model(x)
  File "/home/h/hvaidya/.conda/envs/ocl_survey/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1518, in _wrapped_call_impl
    return self._call_impl(*args, **kwargs)
  File "/home/h/hvaidya/.conda/envs/ocl_survey/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1527, in _call_impl
    return forward_call(*args, **kwargs)
  File "/home/h/hvaidya/.conda/envs/ocl_survey/lib/python3.10/site-packages/avalanche/models/dynamic_modules.py", line 460, in forward
    return self.eval_classifier(x)
  File "/home/h/hvaidya/.conda/envs/ocl_survey/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1518, in _wrapped_call_impl
    return self._call_impl(*args, **kwargs)
  File "/home/h/hvaidya/.conda/envs/ocl_survey/lib/python3.10/site-packages/torch/nn/modules/module.py", line 1527, in _call_impl
    return forward_call(*args, **kwargs)
  File "/home/h/hvaidya/.conda/envs/ocl_survey/lib/python3.10/site-packages/torch/utils/_contextlib.py", line 115, in decorate_context
    return func(*args, **kwargs)
  File "/home/h/hvaidya/.conda/envs/ocl_survey/lib/python3.10/site-packages/avalanche/models/ncm_classifier.py", line 77, in forward
    assert self.class_means_dict != {}, "no class means available."
AssertionError: no class means available.
  0%|▎
AlbinSou commented 1 year ago

ICarl is not mean to be run with use_task_boundaries=False . It needs to access the task boundaries during training in order to know when to store the old model. Furthermore, we did not include icarl in the survey because we were not able to reach satisfying results with it. This kind of bug was happening before even with use_task_boundaries=True but we fixed it. I don't know what is causing it this time but I don't think this requires a fix since the strategy is not meant to be used that way.