EMI | ICML 2019 paper , EMI : Exploration with Mutual Information

You can use nth-child range to filter for columns 1 and 2 of table along with the appropriate row offset within the table initially matched by class.

Then extract url or date (as text) within list comprehensions over split initial returned list (as will alternate column 1 column 2 column 1 etc). Complete the url or convert to actual dates (text) within respective list comprehensions, zip the resultant lists and convert to DataFrame

The data you see is loaded from external source via JavaScript. You can use their API to get the data. For example:

Select your elements more specific e.g. with css selectors and be aware you have to concat the href with baseUrl:

I'm not really clear on what you're after here. In particular, your code apparently doesn't always do what you want it to do ("it works perfectly for the first two words, but not the third"), but you haven't asked for a solution to that, nor given us enough information to know why the third word "is wrong".

So I'll just make stuff up ;-) Since re.sub() doesn't know about overlaps, I'd match multiple times in "priority" order, looking only for things of the form

You can do:

Since you didn't share a link to that page I can only guess.
So, instead of

ftplib docs suggest to use address without leading ftp:// I did

Since you want to dedup within the catalog, I would use generate-id() and create a composite key with the generated ID for the catalog element and the cd/title.

Use that composite key for the xsl:key matching on cd, then you can iterate over each /hello/catalog and then for each cd under that catalog you can use the key for Muenchian grouping:

I was able to get about 25 % faster by combining the three processes.