Reproduce the tf-binding, but "sh: 1: gkmpredict: Permission denied" has always occurred, I don't want to give up!

[Licko0909]

I am using the tutorial to execute snakemake to reproduce the tf-binding. So far, the .h5 files of the four models have been obtained. Still the last step!

But the last lsgkm-SVM model has failed, the results are as follows:
sh: 1: gkmpredict: Permission denied

This is a screenshot of the same stage when DeepSEA is running. It can be compared with the above figure.

I have been unable to find the problem. I have followed the tutorial and it has been going on for many days. I am a college student and really want to reproduce and learn from it, I really don't want to give up!

One final question: Kipoi does not seem to support users retraining existing models with their own data?

[Licko0909]

Later, I tried to run lsgkm-SVM separately, and it runs successfully, but it takes 45 hours. Is it what I did wrong or does it really take so long?

I read the paper mentioned in the paper and found that SVM seems to take the longest time.

[Avsecz]

It really takes that long (e.g. 100x slower than other methods)... I had to let it run for a few days.

[Licko0909]

It really takes that long (e.g. 100x slower than other methods)... I had to let it run for a few days.

Thank you very much!
After I asked, I solved this problem. I later changed the kipoi-manuscript environment to lskg-SVM environment to solve sh: 1: , it only takes 9 hours, but it will not run after only 12 minutes. This is how I use top to view it. Confirmed after the cpu operation. Later, I tried it a few times and only interrupted it after running for 12 minutes. This is very strange!

[Licko0909]

It really takes that long (e.g. 100x slower than other methods)... I had to let it run for a few days.

One final question: Kipoi does not seem to support users retraining existing models with their own data?

[Avsecz]

Kipoi doesn't provide any specific code for retraining existing models, but it's fairly easy to do if you are familiar with the framework in which the model was written in (e.g. Keras). You can directly use the data-loader for training the model (and use multiple workers to load the data efficiently). See https://github.com/kipoi/manuscript/blob/master/src/transfer_learning/tlearn.py for an example how to do transfer learning with a kipoi model written in Keras.

[Avsecz]

If you want to train a model from scratch using the exact same architecture you can use the following code to randomly re-initialize model parameters: keras-team/keras#341 (comment)

[Licko0909]

It really takes that long (e.g. 100x slower than other methods)... I had to let it run for a few days.

Thank you very much!
After I asked, I solved this problem. I later changed the kipoi-manuscript environment to lskg-SVM environment to solve sh: 1: , it only takes 9 hours, but it will not run after only 12 minutes. This is how I use top to view it. Confirmed after the cpu operation. Later, I tried it a few times and only interrupted it after running for 12 minutes. This is very strange!

I can't get the .h5 file at the end.