AttributeError: Can‘t pickle local object ‘get_scheduler.＜locals＞.＜lambda＞‘

pytorch lightning设置优化器的学习率调整策略为LambdaLR时，训练一轮后出现如下报错

lib/python3.7/site-packages/lightning_lite/plugins/io/torch_io.py:61: UserWarning: Warning, `hyper_parameters` dropped from checkpoint. Ais not picklable: Can't pickle local object 'get_scheduler.<locals>.<lambda>'
  rank_zero_warn(f"Warning, `{key}` dropped from checkpoint. An attribute is not picklable: {err}")
Traceback (most recent call last):
  File "/home/server-12/anaconda3/envs/casmvsnet/lib/python3.7/site-packages/lightning_lite/plugins/io/torch_io.py", line 54, in save_checkpoint
    _atomic_save(checkpoint, path)
  File "/home/server-12/anaconda3/envs/casmvsnet/lib/python3.7/site-packages/lightning_lite/utilities/cloud_io.py", line 67, in _atomic_save
    torch.save(checkpoint, bytesbuffer)
  File "/home/server-12/anaconda3/envs/casmvsnet/lib/python3.7/site-packages/torch/serialization.py", line 379, in save
    _save(obj, opened_zipfile, pickle_module, pickle_protocol)
  File "/home/server-12/anaconda3/envs/casmvsnet/lib/python3.7/site-packages/torch/serialization.py", line 589, in _save
    pickler.dump(obj)
AttributeError: Can't pickle local object 'get_scheduler.<locals>.<lambda>'

问题解决：

将代码的匿名函数传入改为用变量存储的具名函数传入

scheduler = LambdaLR(optimizer, lambda epoch: (1-epoch/hparams.num_epochs)**hparams.poly_exp)

改为

lambda1 = lambda epoch: (1-epoch/hparams.num_epochs)**hparams.poly_exp
scheduler = LambdaLR(optimizer, lambda1)

问题原因：

pickle无法处理不带有名字的lambda匿名返回值，需要用变量存储后再传入

注：

如果还不行，如果你的网络设置了warmup，把warmup epoch次数设置为0试一试