TypeError: __init__() got an unexpected keyword argument 'checkpoint_callback'

[wershofe]

Hi
I just tried running the Norman data tutorial and get the following error:

model.train(max_epochs=2000,
use_gpu=True,
batch_size=1024,
early_stopping=True,
plan_kwargs=trainer_params,
early_stopping_patience=15,
check_val_every_n_epoch=5,
save_path='/home/mohsen/projects/cpa/lightning_logs/Norman2019_prep_new/',
)

TypeError Traceback (most recent call last)
Cell In [16], line 1
----> 1 model.train(max_epochs=2000,
2 use_gpu=True,
3 batch_size=1024,
4 early_stopping=True,
5 plan_kwargs=trainer_params,
6 early_stopping_patience=15,
7 check_val_every_n_epoch=5,
8 save_path='/home/mohsen/projects/cpa/lightning_logs/Norman2019_prep_new/',
9 )

File ~/miniconda3/envs/envCPA/lib/python3.9/site-packages/cpa/_model.py:317, in CPA.train(self, max_epochs, use_gpu, train_size, validation_size, batch_size, early_stopping, plan_kwargs, hyperopt, save_path, **trainer_kwargs)
314 checkpoint = SaveBestState(monitor='cpa_metric', mode='max', period=20, verbose=False)
315 trainer_kwargs['callbacks'].append(checkpoint)
--> 317 runner = TrainRunner(
318 self,
319 training_plan=self.training_plan,
320 data_splitter=data_splitter,
321 max_epochs=max_epochs,
322 use_gpu=use_gpu,
323 early_stopping_monitor="cpa_metric",
324 early_stopping_mode='max',
325 checkpoint_callback=True,
326 **trainer_kwargs,
327 )
328 runner()
330 self.epoch_history = pd.DataFrame().from_dict(self.training_plan.epoch_history)

File ~/miniconda3/envs/envCPA/lib/python3.9/site-packages/scvi/train/_trainrunner.py:67, in TrainRunner.init(self, model, training_plan, data_splitter, max_epochs, use_gpu, **trainer_kwargs)
65 self.lightning_devices = lightning_devices
66 self.device = device
---> 67 self.trainer = Trainer(
68 max_epochs=max_epochs,
69 accelerator=accelerator,
70 devices=lightning_devices,
71 gpus=None,
72 **trainer_kwargs,
73 )

File ~/miniconda3/envs/envCPA/lib/python3.9/site-packages/scvi/train/_trainer.py:141, in Trainer.init(self, gpus, benchmark, flush_logs_every_n_steps, check_val_every_n_epoch, max_epochs, default_root_dir, enable_checkpointing, num_sanity_val_steps, enable_model_summary, early_stopping, early_stopping_monitor, early_stopping_min_delta, early_stopping_patience, early_stopping_mode, enable_progress_bar, progress_bar_refresh_rate, simple_progress_bar, logger, log_every_n_steps, replace_sampler_ddp, **kwargs)
138 if logger is None:
139 logger = SimpleLogger()
--> 141 super().init(
142 gpus=gpus,
143 benchmark=benchmark,
144 check_val_every_n_epoch=check_val_every_n_epoch,
145 max_epochs=max_epochs,
146 default_root_dir=default_root_dir,
147 enable_checkpointing=enable_checkpointing,
148 num_sanity_val_steps=num_sanity_val_steps,
149 enable_model_summary=enable_model_summary,
150 logger=logger,
151 log_every_n_steps=log_every_n_steps,
152 replace_sampler_ddp=replace_sampler_ddp,
153 enable_progress_bar=enable_progress_bar,
154 **kwargs,
155 )

File ~/miniconda3/envs/envCPA/lib/python3.9/site-packages/pytorch_lightning/utilities/argparse.py:345, in _defaults_from_env_vars..insert_env_defaults(self, *args, **kwargs)
342 kwargs = dict(list(env_variables.items()) + list(kwargs.items()))
344 # all args were already moved to kwargs
--> 345 return fn(self, **kwargs)

TypeError: init() got an unexpected keyword argument 'checkpoint_callback'

Are you able to advise how to solve?

Thanks
Esther

[yanwu2014]

I'm also seeing this error when I try to run CPA with a CPU, doesn't seem to happen with a GPU

[yanwu2014]

Ok I think I figured it out: the latest versions of scvi-tools and pytorch-lightning have an updated API which is breaking

I was able to get CPA to run by installing the earliest version of scvi-tools required:
pip3 install scvi-tools==0.16.4

[PierreBoyeau]

Another working solution for me was to pin python to be 3.10.
It seems like the issue is that the package requires an up-to-date lightning version, which requires python>=3.8