Installation issues

When installing all requirements from requirements.txt, there is afterward an error with yaml package (No module named yaml)

This issue might be helpful: Try installing yaml manually with a specific version.

I get an error when installing PyTorch (could not find a version that satisfies the requirement torch)

Depending on the Python version and virtual environment, users may experience errors when installing PyTorch via pip. The most common reason for this problem is that the Python version is too new to be compatible with the torch package. Currently, the torch package on pypi is only supported up to Python version 3.7. We recommend trying to use Python version 3.7 or version 3.8 in a conda virtual environment.

If this does not resolve the problem, try installing PyTorch manually.

On Windows for instance, you can try the following to install PyTorch 1.7.1 for CPU only using pip:

pip install torch==1.7.1+cpu -f

For more information on PyTorch installation for different operating systems, please see the PyTorch documentation, and afterwards try to install immuneML again.

Please note that when using DeepRC from immuneML, a PyTorch distribution that supports GPUs is required.

Runtime issues

When running the TrainMLModel instruction multiple times, sometimes it fails saying that there is only one class in the data. Why does this happen?

Please check the number of examples used for machine learning (e.g. number of repertoires or receptors). If there are very few examples, and/or if classes are not balanced, it is possible that just by chance, the data from only one class will be in the training set. If that happens, the classifiers will not train and an error will be thrown. To fix this, try working with a larger dataset or check how TrainMLModel is specified. If TrainMLModel does nested cross-validation, it might require a bit more data. To perform only cross-validation, under selection key, specify that split_strategy is random and that training_percentage is 1 (to use all data from the inner loop for training). In this way, instead of having multiple training/validation/test splits, there will be only training/test splits as specified under key assessment in TrainMLModel instruction.

When running DeepRC I get TypeError: can’t concat str to bytes

This error occurs when h5py version 3 or higher is used. Try using version 2.10.0 or lower.

I am trying to run a report, but it gives no results and no errors. What happened?

To ensure that large analyses do not crash if one of the reports failed (for example, if an error occurs during calculating the results or plotting), reports are run in a ‘safe mode’ so that errors do not stop the execution. Check the output log.txt file to see if any errors or warnings were produced by the reports you tried to run.