Parameters of Model Loading before Inference

[rose4labour]

Below is the model loading code, for the code “class_imbalance_ratio = train_ratios,“ My test indicates that loading the model requires specifying the train_ratios. I only need inference without a test csv file, so I specify 138 values fo train_ratios, like [0.1,0.2,...] I've tested it several times, and different 138 values of train_ratios seem to be independent of the inference results. Is this correct? if I only need to do inference, which part of parameters can I modify for a specific model weight file (such as https://github.com/ARY2260/openpom/blob/main/examples/example_model.pt)? THANK U!

model = MPNNPOMModel(n_tasks=n_tasks, batch_size=128, learning_rate=learning_rate, class_imbalance_ratio = train_ratios, loss_aggr_type = 'sum', node_out_feats = 100, edge_hidden_feats = 75, edge_out_feats = 100, num_step_message_passing = 5, mpnn_residual = True, message_aggregator_type = 'sum', mode = 'classification', number_atom_features = GraphConvConstants.ATOM_FDIM, number_bond_features = GraphConvConstants.BOND_FDIM, n_classes = 1, readout_type = 'set2set', num_step_set2set = 3, num_layer_set2set = 2, ffn_hidden_list= [392, 392], ffn_embeddings = 256, ffn_activation = 'relu', ffn_dropout_p = 0.12, ffn_dropout_at_input_no_act = False, weight_decay = 1e-5, self_loop = False, optimizer_name = 'adam', log_frequency = 32,

model_dir = f'./example_model.pt',

                    device_name='cuda')

[ARY2260]

Yes, you are correct. Imbalance ratios are used only by the loss function during training of the model. It doesn't have any impact on inferences since its only a forward pass.

Thank you for reporting this bug!

For the work around during inference, you can use an arbitrary n_tasks size array. The value n_tasks should be the same as the value used during training of the model.

[rose4labour]

Yes, you are correct. Imbalance ratios are used only by the loss function during training of the model. It doesn't have any impact on inferences since its only a forward pass.

Thank you for reporting this bug!

For the work around during inference, you can use an arbitrary n_tasks size array. The value n_tasks should be the same as the value used during training of the model.

Thanks for your answer. What about the best training parameters? For example, num_step_message_passing = 5 what about num_step_message_passing =4 or 6?