allchemist

both regression and classification tasks, with use mean_squared_error and sigmoid_cross_entropy loss functions. Do i need to modify my loss functions for this trick? or i can provide a modified metric...

I suppose metrics do not participate in backward pass, so i have to modify loss functions

Weighted loss function is an easy part. Seems like sending only two arguments to loss function is hard-coded inside models/prediction/regressor.py and classifier.py, and also in `GraphConvPredictor` definition (which is not...

Thanks for answer, i found out that manual setting values to train_dataset.w does not really change them (after assigning train_dataset.w[0,0] = 0 it actually stays 1), so my experiment was...

By the way, can you share some ideas why multitask learning with missed values does not work? I tried different models and different datasets, training one data column works fine,...

I found the problem - in ValidationCallback `model.evaluate` method is used, which did not use weights from validation dataset. The `use_sample_weights` flag in `model.evaluate` returns error, so i wrote a...

Also, if i do not want to use @route syntax, what function call should i do instead?