What could be compared against for prediction result?

[jtxtina]

Since there is no comparison in prediction part, I am wondering is there anything that we can adopt to compare with to get accuracy result or something similar(Like in training part)? To better evaluate the model.

[andrewkern]

Hi there-- if you know the true labels from a set of observations, then you can use the prediction mode with a trained classifier to assess accuracy on that set.

Does that help?

[jtxtina]

Yeah. But, do you have any such observations or test datasets?

[jtxtina]

Btw, have you ever trained with human gene datasets? Like what you did in S/HIC.

[jtxtina]

I am currently using Tennesse Euro training set in SHIC to train here. For calculating feature vector from simulation step, I am troubled by masking file. Could you help me with this? Thank you!

[jtxtina]

I have finished a successful run of training, but the val_accuracy is really low, around 0.38, anything could helpful?

[andrewkern]

Yeah. But, do you have any such observations or test datasets?

you should use simulated data for this

[andrewkern]

Btw, have you ever trained with human gene datasets? Like what you did in S/HIC.

i'm not sure what you mean.

[andrewkern]

I am currently using Tennesse Euro training set in SHIC to train here. For calculating feature vector from simulation step, I am troubled by masking file. Could you help me with this? Thank you!

The masking file is a fasta formated file with N's in place of the bases you wish to mask out

[jtxtina]

Yeah. But, do you have any such observations or test datasets?

you should use simulated data for this

Well, I mean, if there is any real dataset for this?