THEMIS | Tumor Heterogeneity Extensible Modeling | Topic Modeling library

Looks like you have things pretty well along!

If I understand correctly, you are trying to capture only one # symbol at the beginning of each line.

The regular expression that helps you solve the issue is: r"(?:^|\s)(?:[#]\ )(.*\n+##\ ([^#]*\n)+)". The brackets isolate the capturing or non capturing groups. The first group (?:^|\s) is a non capturing group, because it starts with a question mark. Here you want that your matched string starts with the beginning of a line or a whitespace, then in the second group ([#]\ ), [#] will match exactly one # character. \ matches the space between the hash and the h1 tag text content. finally you want to match any possible character until the end of the line so you use the special characther ., which identifies any character, followed by + that will match any repetition of the previous matched character.

This is probably the code snippet you are looking for, I tested it with the same sample test you used.

You may get not all content back with requests, cause it is served dynamically by the website, but you can use selenium to fix that.

Example

I think a lot of what you did, specifically around data wrangling, was not necessary, especially since you called d3.hierarchy() and d3.cluster() afterwards. I've replaced this with d3.stratify (which deals with hierarchical data that is not yet in the right format).

I've also replaced d3.cluster with d3.tree() because it was unclear to me why you'd want to use d3.cluster here. Your data has multiple parents, multiple roots and even floating nodes, and d3 is not meant to deal with that. My workaround has been to attach pseudonodes to every level, so as to make sure that there is only one node and that all nodes are at the right level at all times. To make sure the links were drawn correctly, I've written a custom getLinks function, that can deal with multiple parents.

I've also written a custom link generator that draws the links somewhat in the way that you want them. d3 doesn't offer much of flexibility here, but you can use the source code for inspiration.

Edit

I've changed the logic to be more focused on which "partners" got a child, so both links to the same child are on the same level - like in your picture. I've also drawn the nodes based on how many partners they have, and have given every link an offset so the lines are more distinct.

I've sorted the nodes so that the real pro-creators are at the top (Zeus), which gives a more balanced and less crowded view.

If you have a specific range for num_comp that you want to try out, I wouldn't bother with going to the workflow and getting the parameters, etc. I would set up the tuning grid with the parameters directly:

Try to convert your predicted probabilities to labels, and then run your confusionMatrix on this:

Problem was solved by modifying cacheLevelName to CACHE_NONE instead of CACHE_CONSUMER.

This seems to work. It assumes that

• The first line that contains numbers is that immediately after the line that begins with 'dd-mm-yyyy'.
• The last line that contains numbers is two lines above the line that begins with '# Key Parameter'.

Code:

The problem is that spring changes the return content-type if the end of a URL has an extension.

So seeing .zip at the end, was causing spring to over-ride the type to application/zip .