geschichte | zustand and immer based hook to manage query parameters

I downloaded the Wikipedia data in smaller chunks from here. I unzipped the files and now I want to extract the text from them (the largest are over 3 GB). I have a code that works, but it crashes when the file is too large:

I am looking for a way to access the li-elements between two specific heading-tags only (e.g.from 2nd h3 to 3rd h3 or from 3rd h3 to next h4) in order to create a table of historical events listed on https://de.wikipedia.org/wiki/1._Januar structured along the criteria mentioned in the headings. A major problem (for me ...) is that - other than the h1-heading - the subtitles of the lower levels have no className or id.

I have seen several similar questions, but none that addressed specifically my problem:

given a novel in xml file (this is a very small cut from the start and the end)

In an earlier question (Teacher-Student System: Training Student With k Target Sequences for Each Input Sequence), I wanted a teacher machine translation (MT) model to perform an online search during the training of a student speech translation (ST) model, to generate multiple targets per input sequence for the student.

Now, in the hope to speed things up, I want the teacher to perform the search offline. So, I let the teacher generate a hypothesis file containing the output of its ChoiceLayer.

My plan is to use these hypotheses and the ground truth as target for the student. So, for each input sequence from source.train(.gz) there is the ground truth target coming from target.train(.gz) and one or multiple teacher hypotheses coming from teacher.train.hyp.

In my online search config, I simply registered the teacher’s Data coming from its ChoiceLayer as extern_data. Since the Data contained a SearchBeam, RETURNN automatically regarded that the target Data consists of multiple sequences (this feature was added with issue Targeting Beam as Training Target Causes Dimension Error #304). I want a similar automatism in my offline search solution.

My question is, does RETURNN already have components, with which my plan could be achieved, or do I have to write an extension for that? If I have to extend RETURNN for that, does anybody have recommendations on how to do that? (E.g. I thought about writing a special TranslationDataset subclass)

EDIT (15.8.2020):

First, let me answer Albert's question, on how my data currently looks like (which is subject to change, of course):

Currently, I don't make use of HDF files or HDFDataset. I use MetaDataset to combine an ExternSprintDataset and TranslationDataset to combine the ASR and the MT (ground truth) data for my ST student, but yet I'm not at the point where I load the teacher's hypotheses from any persistent data format.

However, I've dumped the teacher's hypotheses by simply selecting its ChoiceLayer (decoder output) as search_output_layer. Then, on search, RETURNN creates a text file (in my case I chose teacher.train.hyp), which contains the hypotheses per input sequence stored, basically as a string-serialized Python-dict, like this:

I'm new to coding, and for my first project I wanted to do something useful and wrote a code that calculates your grade at the end of the year (only works in Germany).

My problem is that some students have 15 classes and some 16, and I want to include an if option, but I don't know how. Could someone help me?

Here is the code:

We are validating our XSD through https://www.freeformatter.com/xml-validator-xsd.html but it throws an error:

S4s-elt-must-match.1: The Content Of 'filmliste' Must Match (annotation?, (simpleType | ComplexType)?, (unique | Key | Keyref)*)). A Problem Was Found Starting At: Sequence.

Can someone help us?

Below is our XML and XSD Code (We changed the schemaLocation in the XML to XXXX just for the code preview):

I am working on a website, only the mobile view is somewhat complete.

I want to have the scroll to section functionality and can make it work if I assign a separate function and event listener to the individual links, but that would be repetitive.

The markup for the menu is:

My Input file its a plain content:

Fries Scheepvaartmuseum: Schiffmodelle in jeglichen Größen und viele Infos über Schiffsbau und Seefahrt sowie über die Geschichte der Stadt Sneek. *www.friesscheepvaartmuseum.nl** Museen sowie facebook.com viele kleine Gassen zwischen den https://facebook.com Grachten locken zu Erkundungstouren. Der Strand lädt zu romantischen Spaziergängen ein https://stackoverflow.com/questions/tagged/perl nicht nur probieren und kaufen, sondern auch das nostalgische Haus und die Destillerie besichtigen stackoverflow.com/questions/tagged/perl

I can able to find www.., https?://. with prefix (www, http) and suffix (list of domains).

However, I need to find the links based on the some list of domains like ... .edu, .com, .af, .ag, .ai, .al without prefix and suffix in the web links.

For example:

I couldn't able to find incomplete or without prefix www, https, http links like facebook.com, stackoverflow.com/questions/tagged/perl in a plain contents.

Could you please someone help me on this one if there is any module is available or any regex patterns would be helpful for me since I have more than 10k web links to find.