You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Leaving the property on the PropertyFileGraph will: a) force a rerun of the algorithm to use a different property name to avoid conflicts, b) waste memory since the old property is referenced and will hang around in memory. These are both issue in a Jupyter notebook where having a failure in a given cell is usually handled by only rerunning that one cell.
I think a good option is adding a rollback (delete the new property) if the algorithm fails. Another options would be to chang the property creation semantics of algorithms to be "add or overwrite", but I think that could make it easy to accidentally destroy your data due to the silent overwrite.
Fixing this should be a relatively straightforward matter. Many algorithms already have a thin wrapper function that could easily be modified to check for a returned error and delete the output property if it exists. Other algorithms could have such a thin wrapper added if they have a large number of returns making removing the property in every care messy.
Leaving the property on the PropertyFileGraph will: a) force a rerun of the algorithm to use a different property name to avoid conflicts, b) waste memory since the old property is referenced and will hang around in memory. These are both issue in a Jupyter notebook where having a failure in a given cell is usually handled by only rerunning that one cell.
I think a good option is adding a rollback (delete the new property) if the algorithm fails. Another options would be to chang the property creation semantics of algorithms to be "add or overwrite", but I think that could make it easy to accidentally destroy your data due to the silent overwrite.
This originally came up in #5 (comment).
The text was updated successfully, but these errors were encountered: