TypeError: Object of type Tensor is not JSON serializable when running train_transformer_style with takes_target as 1

[97harsh]

Traceback (most recent call last):
File "flood_forecast/trainer.py", line 108, in
main()
File "flood_forecast/trainer.py", line 103, in main
train_function(training_config["model_type"], training_config)
File "flood_forecast/trainer.py", line 42, in train_function
train_transformer_style(model=trained_model,
File "/home/harsh/Documents/Coronawhy/flow-forecast/flood_forecast/pytorch_training.py", line 146, in train_transformer_style
model.save_model(model_filepath, max_epochs)
File "/home/harsh/Documents/Coronawhy/flow-forecast/flood_forecast/time_model.py", line 152, in save_model
json.dump(self.params, p)
File "/home/harsh/anaconda3/envs/flow-forecast/lib/python3.8/json/init.py", line 179, in dump
for chunk in iterable:
File "/home/harsh/anaconda3/envs/flow-forecast/lib/python3.8/json/encoder.py", line 431, in _iterencode
yield from _iterencode_dict(o, _current_indent_level)
File "/home/harsh/anaconda3/envs/flow-forecast/lib/python3.8/json/encoder.py", line 405, in _iterencode_dict
yield from chunks
File "/home/harsh/anaconda3/envs/flow-forecast/lib/python3.8/json/encoder.py", line 405, in _iterencode_dict
yield from chunks
File "/home/harsh/anaconda3/envs/flow-forecast/lib/python3.8/json/encoder.py", line 438, in _iterencode
o = _default(o)
File "/home/harsh/anaconda3/envs/flow-forecast/lib/python3.8/json/encoder.py", line 179, in default
raise TypeError(f'Object of type {o.class.name} '
TypeError: Object of type Tensor is not JSON serializable

[97harsh]

flow-forecast/flood_forecast/pytorch_training.py

Lines 211 to 212 in 7927c51

	if takes_target:
	forward_params["t"] = trg

this is updating the forward_params with Tensor objects in models.params when takes_target==1
This should be sent as copy of dictionary instead of dictionary itself

flow-forecast/flood_forecast/pytorch_training.py

Lines 90 to 98 in 7927c51

	total_loss = torch_single_train(
	model,
	opt,
	criterion,
	data_loader,
	takes_target,
	meta_model,
	meta_representation,
	forward_params)

[97harsh]

@isaacmg should I create a PR to should I change this ?

[isaacmg]

This didn't seem to occur before with the full transformer though

flow-forecast/tests/full_transformer.json

Line 54 in 7927c51

"t":{}

[isaacmg]

If that is the sole issue one would think this test should fail too

flow-forecast/.circleci/config.yml

Line 251 in 7927c51

echo -e 'running trainer_full_transformer_test \n'

[97harsh]

The value t only gets tensor object when takes_target argument is passed as True to torch_single_train