Troubleshooting

NoSuchKernel Errors (using Conda)

NoSuchKernel Errors can appear when running papermill on jupyter notebooks whose kernel has been specified via conda (nb_conda). nb_conda is used to easily set conda environment per notebook from within jupyterlab.

To illustrate, the following example demonstrates the creation of a new environment with all the dependencies necessary for an analysis.

conda create -n analysis_1 python=2 ipykernel

Once nb_conda is used within the jupyter server to set the kernel for a notebook to analysis_1, the notebook gets metadata similar to the following:

{
 "kernelspec": {
  "display_name": "Python [conda env:analysis_1]",
  "language": "python",
  "name": "conda-env-analysis_1-py"
 }
}

Papermill cannot use this metadata to determine that it should use analysis_1 to execute this notebook. Running papermill (from analysis_1 or another environment) will raise the following error:

jupyter_client.kernelspec.NoSuchKernel: No such kernel named conda-env-analysis_1-py

This can be fixed by:

• Installing jupyter (or at least ipykernel) in analysis_1

conda install -n analysis_1 jupyter

• Expose the analysis_1 environment as a jupyter kernel (this is no longer automatic).

conda activate analysis_1
jupyter kernelspec install --user --name analysis_1

• Run papermill (from any environment) specifying the correct kernel using the -k option

papermill my_notebook.ipynb output_notebook.ipynb -k analysis_1