config parsing error?

[KyonP]

Hello, I am a newbie who recently started working on a project based on GAN with Pytorch-lightning.

I managed to build a primitively long code that worked fine; however, I have faced error messages that cannot be resolved since the start of this month.

I spent almost a week figuring out where to look; I hope to get some advice.

here is my error message:

Traceback (most recent call last):
  File "/home/kp/LFS/main.py", line 97, in <module>
    cli_main()
  File "/home/kp/LFS/main.py", line 88, in cli_main
    cli = VlpLightningCLI(Learner,
  File "/usr/local/lib/python3.8/dist-packages/pytorch_lightning/utilities/cli.py", line 516, in __init__
    self.instantiate_classes()
  File "/usr/local/lib/python3.8/dist-packages/pytorch_lightning/utilities/cli.py", line 653, in instantiate_classes
    self.config_init = self.parser.instantiate_classes(self.config)
  File "/usr/local/lib/python3.8/dist-packages/jsonargparse/deprecated.py", line 127, in patched_instantiate_classes
    cfg = self._unpatched_instantiate_classes(cfg, **kwargs)
  File "/usr/local/lib/python3.8/dist-packages/jsonargparse/core.py", line 1067, in instantiate_classes
    cfg[subcommand] = subparser.instantiate_classes(cfg[subcommand], instantiate_groups=instantiate_groups)
  File "/usr/local/lib/python3.8/dist-packages/jsonargparse/deprecated.py", line 127, in patched_instantiate_classes
    cfg = self._unpatched_instantiate_classes(cfg, **kwargs)
  File "/usr/local/lib/python3.8/dist-packages/jsonargparse/core.py", line 1062, in instantiate_classes
    component.instantiate_class(component, cfg)
  File "/usr/local/lib/python3.8/dist-packages/jsonargparse/signatures.py", line 555, in group_instantiate_class
    parent[key] = group.group_class(**value)
  File "/home/kp/LFS/lfs/modules/learner.py", line 102, in __init__
    self.fid = EfficientFrechetInceptionDistance()
  File "/home/kp/LFS/lfs/utils/fid.py", line 29, in __init__
    super().__init__(
TypeError: __init__() takes from 1 to 3 positional arguments but 6 were given

Process finished with exit code 1

I am familiar with the error message itself. However, it is hard to distinguish the problem source.

The last reached point self.fid = EfficientFrechetInceptionDistance() is a code that inherited from torchmetrics.image.fid :

from typing import Any, Callable, List, Optional, Union

import numpy as np
import torch
import warnings
from deprecate import deprecated, void
from torch import Tensor
from torch.autograd import Function

from torchmetrics.metric import Metric
from torchmetrics.image.fid import FrechetInceptionDistance, _compute_fid
from torchmetrics.utilities import _future_warning, rank_zero_info, rank_zero_warn
from torchmetrics.utilities.data import dim_zero_cat
from torchmetrics.utilities.imports import _SCIPY_AVAILABLE, _TORCH_FIDELITY_AVAILABLE


class EfficientFrechetInceptionDistance(FrechetInceptionDistance):
    def __init__(
        self,
        feature: Union[int, torch.nn.Module] = 2048,
        limit: int = 60000,
        compute_on_step: bool = False,
        dist_sync_on_step: bool = False,
        process_group: Optional[Any] = None,
        dist_sync_fn: Callable[[Tensor], List[Tensor]] = None,
    ) -> None:
        warnings.filterwarnings("ignore",
            "Metric `FrechetInceptionDistance` will save all extracted")
        super().__init__(
            feature,
            compute_on_step,
            dist_sync_on_step,
            process_group,
            dist_sync_fn,
        )

        self.limit = limit
        self.add_state("real_mean", torch.zeros(feature), dist_reduce_fx="sum")
        self.add_state("real_cov", torch.zeros(
            feature, feature), dist_reduce_fx="sum")
        self.add_state("real_count", torch.tensor(0), dist_reduce_fx="sum")
        self.add_state("fake_mean", torch.zeros(feature), dist_reduce_fx="sum")
        self.add_state("fake_cov", torch.zeros(
            feature, feature), dist_reduce_fx="sum")
        self.add_state("fake_count", torch.tensor(0), dist_reduce_fx="sum")

        def remove_state(key):
            del self._defaults[key]
            del self._persistent[key]
            del self._reductions[key]
        remove_state("real_features")
        remove_state("fake_features")

    def update(self, imgs: Tensor, real: bool, inception: bool = False) \
    -> None:  # type: ignore
        """Update the state with extracted features.
        Args:
            imgs: tensor with images feed to the feature extractor
            real: bool indicating if imgs belong to the real or the fake distribution
            inception: bool indicating if imgs are inception features
        """
        x = self.inception(imgs) if not inception else imgs
        self.orig_dtype = x.dtype
        x = x.double()

        if real:
            self.real_count += x.shape[0]
            self.real_mean += x.sum(dim=0)
            self.real_cov += x.t().mm(x)
        else:
            excess = self.fake_count + x.shape[0] - self.limit
            if 0 < excess:
                x = x[:-excess]
            if 0 < x.shape[0]:
                self.fake_count += x.shape[0]
                self.fake_mean += x.sum(dim=0)
                self.fake_cov += x.t().mm(x)

    def compute(self) -> Tensor:
        """Compute the FID metric. This is called for each process.
        """
        def outer_product(x):
            return x.unsqueeze(1).mm(x.unsqueeze(0))

        # calculate mean and covariance
        n = self.real_count
        m = self.fake_count
        mean1 = self.real_mean / n
        mean2 = self.fake_mean / m
        cov1 = 1.0 / (n - 1) * (self.real_cov - n * outer_product(mean1))
        cov2 = 1.0 / (m - 1) * (self.fake_cov - m * outer_product(mean2))

        # compute fid
        return _compute_fid(mean1, cov1, mean2, cov2).to(self.orig_dtype)

Since the code worked fine during April, I presumed this occurred by automated or mistakenly conducted Ubuntu updates or Python dependency updates. However, I checked my history, and nothing looks suspicious. FYI, here is my PIP package list:

Package                       Version   
----------------------------- ----------
absl-py                       1.0.0     
aiohttp                       3.8.1     
aiosignal                     1.2.0     
alabaster                     0.7.12    
async-timeout                 4.0.2     
attrs                         21.4.0    
Babel                         2.10.1    
cachetools                    5.0.0     
certifi                       2019.11.28
chardet                       3.0.4     
charset-normalizer            2.0.12    
click                         8.1.3     
clip                          1.0       
dbus-python                   1.2.16    
distro                        1.4.0     
docstring-parser              0.13      
docutils                      0.17.1    
einops                        0.4.1     
ffmpeg                        1.4       
ffmpeg-python                 0.2.0     
filelock                      3.6.0     
frozenlist                    1.3.0     
fsspec                        2022.3.0  
ftfy                          6.1.1     
future                        0.18.2    
google-auth                   2.6.6     
google-auth-oauthlib          0.4.6     
grpcio                        1.46.0    
huggingface-hub               0.5.1     
idna                          2.8       
imagesize                     1.3.0     
importlib-metadata            4.11.3    
Jinja2                        3.1.2     
joblib                        1.1.0     
jsonargparse                  4.1.0     
jsonpatch                     1.32      
jsonpointer                   2.3       
lightning-bolts               0.5.0     
lpips                         0.1.4     
Markdown                      3.3.7     
markdown-it-py                2.1.0     
MarkupSafe                    2.1.1     
mdit-py-plugins               0.3.0     
mdurl                         0.1.1     
multidict                     6.0.2     
myst-parser                   0.17.2    
ninja                         1.10.2.3  
numpy                         1.22.3    
oauthlib                      3.2.0     
opencv-python                 4.5.5.64  
packaging                     21.3      
pandas                        1.4.2     
Pillow                        9.1.0     
pip                           20.0.2    
protobuf                      3.20.1    
pyarrow                       8.0.0     
pyasn1                        0.4.8     
pyasn1-modules                0.2.8     
pyDeprecate                   0.3.1     
Pygments                      2.12.0    
PyGObject                     3.36.0    
pyparsing                     3.0.9     
python-dateutil               2.8.2     
pytorch-lightning             1.5.6     
pytz                          2022.1    
PyYAML                        6.0       
pyzmq                         22.3.0    
regex                         2022.4.24 
requests                      2.22.0    
requests-oauthlib             1.3.1     
rsa                           4.8       
sacremoses                    0.0.53    
scipy                         1.8.0     
setuptools                    45.2.0    
siren-pytorch                 0.1.6     
six                           1.14.0    
snowballstemmer               2.2.0     
Sphinx                        4.5.0     
sphinx-copybutton             0.5.0     
sphinx-favicon                0.2       
sphinx-rtd-theme              1.0.0     
sphinxcontrib-applehelp       1.0.2     
sphinxcontrib-devhelp         1.0.2     
sphinxcontrib-htmlhelp        2.0.0     
sphinxcontrib-jsmath          1.0.1     
sphinxcontrib-katex           0.8.6     
sphinxcontrib-qthelp          1.0.3     
sphinxcontrib-serializinghtml 1.1.5     
ssh-import-id                 5.10      
tensorboard                   2.9.0     
tensorboard-data-server       0.6.1     
tensorboard-plugin-wit        1.8.1     
tokenizers                    0.10.3    
toolz                         0.11.2    
torch                         1.11.0    
torch-fidelity                0.3.0     
torchfile                     0.1.0     
torchmetrics                  0.9.0.dev0
torchvision                   0.12.0    
tornado                       6.1       
tqdm                          4.64.0    
transformers                  4.14.1    
typing-extensions             4.2.0     
urllib3                       1.25.8    
visdom                        0.1.8.9   
wcwidth                       0.2.5     
websocket-client              1.3.2     
Werkzeug                      2.1.2     
wheel                         0.34.2    
yarl                          1.7.2     
zipp                          3.8.0  

I updated and downgraded packages, but nothing changed.

The entry code is main.py.

import os
import time
import resource
import argparse

import torch
from torch.nn import functional as F
import torchvision
import torch.nn as nn

import pytorch_lightning as pl
from pytorch_lightning import LightningModule, seed_everything, Trainer
from pytorch_lightning.loggers import TensorBoardLogger
from pytorch_lightning.callbacks import LearningRateMonitor
from pytorch_lightning.utilities.cli import LightningCLI
from torchmetrics.functional import accuracy
from pytorch_lightning.plugins.training_type import DDPPlugin

from lfs.datamodules import MyDataModule
from lfs.modules import Learner, VideoPredictionWriter
from lfs.datamodules import ImageTextDataModule

class VlpLightningCLI(LightningCLI):
    def add_arguments_to_parser(self, parser):
        parser.link_arguments('data.media_size', 'model.media_size')
        parser.link_arguments('data.batch_size_per_gpu', 'model.batch_size')

        parser.add_argument("--save_validate_imgs", default=False)
        parser.add_argument("--save_test_imgs", default=False)

    def before_instantiate_classes(self):
        if 'fit' in self.config:
            config = self.config['fit']  # changed to this maybe in PL v1.5
            if not isinstance(config['trainer']['gpus'], list): # gpu specifying list check
                if 1 < config['trainer']['gpus']: # DDPPlugin
                    config['trainer']['strategy'] = DDPPlugin(
                        find_unused_parameters=True)
                # inject model label to tensorboard logger
                if hasattr(self.config['fit'], 'model'):
                    label = config['model']['label']
                    config['trainer']['logger'][0]['init_args']['version'] = label
                    config['trainer']['callbacks'][0]['init_args']['filename'] = \
                        'lfs-' + label + '-{epoch:02d}-{val_loss:.2f}'
            elif len(config['trainer']['gpus']) > 1:
                raise Exception('DDP is not supported for specified multiple GPUs')
            else:
                print('single GPU specified : ', config['trainer']['gpus'])
        elif 'predict' in self.config:
            config = self.config['predict']
            if 'cub' in self.config['predict']['data']['media_dir'].lower():
                predictWriter = CubPredictionWriter()
            else:
                predictWriter = VideoPredictionWriter()
            config['trainer']['callbacks'].append(predictWriter)
        else:
            print('warning: no config.')


def prepare_run():
    """Make directories used in the Lightning CLI. It seems the CLI doesn't 
    make the directories for logging and checkpoints.
    """
    os.makedirs('save/checkpoints', exist_ok=True)
    os.makedirs('lightning_logs', exist_ok=True)

    """Increase rlimit for pin memory threads
    """
    rlimit = resource.getrlimit(resource.RLIMIT_NOFILE)
    resource.setrlimit(resource.RLIMIT_NOFILE, (65535, rlimit[1]))


def cli_main():
    cli = VlpLightningCLI(Learner,
                          MyDataModule,
                          save_config_callback=None)

if __name__ == "__main__":
    prepare_run()
    cli_main()

And followed by learner.py since it is a very long code; however, it stops very early, and only partial code is followed.

from typing import *
import os
import math
import copy
from collections import OrderedDict

import torch
from torch import Tensor
import torch.nn as nn
import torch.nn.functional as F
from torch.nn import TransformerEncoder, TransformerEncoderLayer
from torch.nn.functional import mse_loss, cross_entropy
import torchvision
import torchvision.transforms as T

import pytorch_lightning as pl
from pytorch_lightning.utilities.distributed import rank_zero_only
from torchmetrics.image.fid import FrechetInceptionDistance
try:
    from torchvision.transforms import InterpolationMode
    BICUBIC = InterpolationMode.BICUBIC
except ImportError:
    BICUBIC = Image.BICUBIC

from transformers.models.bert.modeling_bert import BertEncoder
from transformers import BertModel
from .fmm_siren import FmmSirenNet

from lfs.modules.stylegan2 import Generator, Discriminator
from lfs.modules.lafite import LafiteGenerator, LafiteDiscriminator
# from lfs.datasets import ImageTextBatch
from lfs.datasets import ImageTextSeqBatch
from lfs.datasets import batch_to_export
from lfs.utils import *

import clip

import lfs.mydataset_CLIP_tokenizer as p_CLIP_tokenizer

class Learner(pl.LightningModule):
    def __init__(self,
                 media_size: int = 128,
                 hidden_size: int = 512,
                 batch_size: int = 4,
                 max_text_len: int = 40,
                 demo_path: str = 'demo/',
                 clip_model: str = 'ViT-B/32',
                 semantic_alignment: bool = True,
                 gamma: float = 10,
                 lamda: float = 20,
                 g_lr: float = 0.0025,
                 d_lr: float = 0.0025,
                 b1: float = .0,
                 b2: float = .99,
                 ema_beta: float = 0.9978,
                 conditional_style=False,
                 perturbation=False,
                 ablation_study=None,
                 previous_condition=True):
        super().__init__()
        self.save_hyperparameters()

        # GAN
        self.generator = LafiteGenerator(size=media_size,
                                         style_dim=hidden_size,
                                         n_mlp=8,
                                         channel_multiplier=2,
                                         conditional_style=conditional_style,
                                         previous_condition=True)
        self.discriminator = LafiteDiscriminator(size=media_size,
                                                 branch_dim=hidden_size,
                                                 channel_multiplier=2)

        self.G_ema = copy.deepcopy(self.generator).eval() \
            if 0 < ema_beta else None
        self.G_ema.requires_grad_(False)
        self.fid = EfficientFrechetInceptionDistance()

Since it is an error with positional arguments, I am conjecturing this is caused by:

1. the argument parser (jsonargparse)
2. mishandled config (on main.py or config file)
3. mishandled Pytorch-lightning CLI

I have been stuck in this issue for almost a week and still don't have a clue. Is there any suggestion for a hint? any small advice would be grateful.

p.s. My Development Environment is Pycharm 2021.3.3, remotely connected to a GPU machine, and controlled as a remote interpreter.

[akihironitta]

@KyonP As the error says, you're giving 6 positional arguments (super(), feature, compute_on_step, dist_sync_on_step, process_group, dist_sync_fn) even though the super class (FrechetInceptionDistance) only accepts from 1 to 3 positional arguments:

        super().__init__(
            feature,
            compute_on_step,
            dist_sync_on_step,
            process_group,
            dist_sync_fn,
        )