notebook
notebook copied to clipboard
probcomp
jupyter/datascience-notebook with probcomp libraries
Following the conda installation instructions leads to the package "decorator" being installed with a version that requires python3, which in turn breaks jupyter (see https://stackoverflow.com/questions/69773657/cannot-run-jupyter-notebook-by-python2-7) This can be fixed by...
[Instructions on dockerhub](https://hub.docker.com/r/probcomp/notebook) contain broken links to images: 
All of the python notebooks seem to be duplicated in our image, see screenshot below. 
The notebook described in #7 should demonstrate finding anomalous or unlikely rows.
The notebook described in #7 should show that it is possible to search by similarity to hypothetical rows. *Warning*: This may generate new bugs in [bayeslite](https://github.com/probcomp/bayeslite) which should be flagged...
The notebook described in #7 should show that using combinations of SQL and BQL-similarity-with-respect-to leads to sensible matches as compared with using just SQL as in #11.
The notebook described in #7 should show that filtering convicts via SQL operators rapidly leads to very few matches. For another example of this see the official Crosscat paper.
The notebook described in #7 should include looking at a dependence heatmap as a diagnostic of whether the models are totally crazy or not.
The notebook described in #7 should support building an ensemble of models based on the data loaded in #8.
The notebook described in #7 should support loading the data into BayesDB.