How to set Model Hyper-Parameters in run_recbole()?

[johnny12150]

If I want to change the model hyper-parameters, is the hyper.test my only option?

[hyp1231]

If you use run_recbole, then you can:

1. Write down your customized hyper-parameters into a yaml file, for example, named as your/path/to.yaml. Then

   from recbole.quick_start import run_recbole
   
   run_recbole(model='BPR', dataset='ml-100k', config_file_list=['your/path/to.yaml'])

2. OR, pass a dict of your customized hyper-parameters directly into run_recbole, for example,

   from recbole.quick_start import run_recbole
   
   my_dict = {
     'train_batch_size': 1234
   }
   run_recbole(model='BPR', dataset='ml-100k', config_dict=my_dict)

Please refer to API Docs for more details about run_recbole.

Besides, hyper.test is designed for Parameter Tuning, but not for set models' hyper-parameters.

[johnny12150]

If I take ItemKNN as an example, I can set K just in YAML to tune the hyper-parameters?

[hyp1231]

If you just want to set a specific value of k, and get the corresponding result, then you can set k just in yaml.
Actually, we will load ItemKNN.yaml automatically to get a default value of k.

RecBole/recbole/properties/model/ItemKNN.yaml

Lines 1 to 2 in f0f2d6b

	k: 100
	shrink: 0.0

If you set the value of k in your custom yaml file, it will override the value in the default yaml file.

But if you want to follow a parameter tuning strategy, and tune k to get a better result automatically. Then it's essential to write down the strategy and range of parameters into hyper.test.