Model Checkpointing

[nilsleh]

Describe the bug In main.py, the enable_checkpointing=False is hard coded. When I pass a lightning ModelCheckpoint to create_trainer() then it throws the error: Trainer was configured with "enable_checkpointing=False" but found "ModelCheckpoint" in callback list.

To Reproduce

from deepforest import main
from pytorch_lightning.callbacks import ModelCheckpoint

checkpoint_callback = ModelCheckpoint(
        dirpath=ckpt_dir,
        save_top_k=1,
        monitor="box_recall",
        mode="max",
        every_n_epochs=1,
    )
m = main.deepforest()
m.use_release()
m.create_trainer(callbacks = [checkpoint_callback])
m.trainer.fit(m)

Environment (please complete the following information):

• OS: Ubuntu
• Python version and environment : venv Python 3.8

[easz]

I had the same issue #301. The fix #304 is not enough.

The hard coded part should be removed (See #298)

[bw4sz]

Okay, let's look into this. We want 1) default to be no checkpoint, 2) allow users to add checkpoint. First step is to make a branch with two tests that fail in the current workflow.

[bw4sz]

Creating a branch for this issue https://github.com/weecology/DeepForest/tree/model_checkpoint

[bw4sz]

Tests pass and merged into main.