FAQs & known issues#

Known issues:#

  • If using the autoreload ipython magic, you might get the following error when saving a KantoData dataset:

    Can't pickle <class 'pykanto.dataset.KantoData'>: it's not the same object as pykanto.dataset.KantoData

    Fix: restart your kernel without the autoreload extension.

  • Ray throws a fork support INFO message when using grpcio==1.44.0

    Fork support is only compatible with the epoll1 and poll polling strategies

    Fix: see ray issue #22518.

  • Ray workers crash:
    If your workers crash and/or your machine freezes it is likely that you have run out of RAM. You could try using fewer cpus when eecuting paralell processes. You can do this by, for example, setting parameters.num_cpus to a number lower than your available CPUS. Example: If your KantoData object is called dataset, then run dataset.parameters.update(num_cpus=1)

System dependencies#

  • hdbscan requires gcc to be installed. If you get an error when hdbscan is being installed, then you might need to install gcc first, e.g. by running sudo apt-get install gcc, and then try installing pykanto again.