evaluation part about hlta HOT 9 CLOSED

1. The output does not really show the topic hierarchy. It only shows the keywords of a topic in each line. You may extract the topic hierarchy using tm.hlta.ExtractTopics and then open the HTML file to see the tree structure of the topic model.

2. It seems that we have only uploaded the code for computing the log-likelihood of test data. The topic coherence code was run by my colleague Peixian Chen. I may not have integrated the code to the repository yet. Perhaps you may ask @clairepchen for sending you the code for computing topic coherence.

from hlta.

Yes, it is possibly due to the data set. For example, on the NIPS dataset with 1k to 10k words, Peixian got a hierarchy with 4 to 6 levels (according to the AAAI-2016 paper).

from hlta.

As you mentioned in your example in the last part we run the tm.hlta.ExtractTopic, so my output is the output of this file.
I think for showing the hierarchy I have to run another file?

Yea then I will contact him,
Can you please let me know the file of logLikelihood also?(there is some file in org.latlab but Im not sure if they are)

Thanks again :)

from hlta.

There should be an HTML file (e.g. sample.html) after running tm.hlta.ExtractTopic. You may open that HTML file in a browser to see the hierarchy.

The step for computing loglikelihood is shown in the Testing section.

from hlta.

I have already opened the outputfile which contains two html file, the output of this html file is the same as the sample html file.
I mean like this:
0.379 node tree hidden-unit module 0.136 analog chip 0.242 channel filter source 0.202 circuit voltage 0.228 firing spike 0.255 synaptic fig_ recurrent 0.278 cortex stimulus 0.293 orientation object
the same as the previous which was available in the output file

from hlta.

You may try to open the HTML file in the base directory but not in that topic_output directory. For example, if you run the Quick Example, you will find the sample.html in the base directory:

├── extracted
│   ├── Chen2016-Latent\ Tree\ Models\ for\ Hierarchical\ Topic.txt
│   ├── Chen2016-Progressive\ EM\ for\ Latent\ Tree\ Models\ and\ Hierarchical.txt
│   ├── Poon2010-Variable\ Selection\ in\ Model-Based\ Clustering\ To\ Do\ or\ To\ Facilitate.txt
│   ├── Poon2013-Model-Based\ Clustering\ of\ High-Dimensional\ Data\ Variablea.txt
│   ├── Poon2017-Clustering\ with\ Multidimensional\ Mixture\ Modelsa.txt
│   ├── Poon2017-Topic\ Browsing\ System\ for\ Research\ Papersa.txt
│   ├── Poon2018-UC-LTM\ Unidimensional\ Clustering\ Using\ Latenta.txt
│   ├── Zhang2017-Latent\ Tree\ Analysis.txt
│   └── liu-n-ecml14.txt
├── fonts
│   ├── glyphicons-halflings-regular.eot
│   ├── glyphicons-halflings-regular.svg
│   ├── glyphicons-halflings-regular.ttf
│   ├── glyphicons-halflings-regular.woff
│   └── glyphicons-halflings-regular.woff2
├── lib
│   ├── 32px.png
│   ├── 40px.png
│   ├── bootstrap.min.css
│   ├── bootstrap.min.js
│   ├── custom.css
│   ├── custom.js
│   ├── ie10-viewport-bug-workaround.css
│   ├── ie10-viewport-bug-workaround.js
│   ├── jquery.min.js
│   ├── jquery.tablesorter.min.js
│   ├── jquery.tablesorter.widgets.js
│   ├── jstree.min.js
│   ├── magnific-popup.css
│   ├── style.min.css
│   ├── tablesorter.css
│   └── throbber.gif
├── model.beforeGlobalEM.bif
├── model.bif
├── pdfs
│   ├── Chen2016-Latent\ Tree\ Models\ for\ Hierarchical\ Topic.pdf
│   ├── Chen2016-Progressive\ EM\ for\ Latent\ Tree\ Models\ and\ Hierarchical.pdf
│   ├── Poon2010-Variable\ Selection\ in\ Model-Based\ Clustering\ To\ Do\ or\ To\ Facilitate.pdf
│   ├── Poon2013-Model-Based\ Clustering\ of\ High-Dimensional\ Data\ Variablea.pdf
│   ├── Poon2017-Clustering\ with\ Multidimensional\ Mixture\ Modelsa.pdf
│   ├── Poon2017-Topic\ Browsing\ System\ for\ Research\ Papersa.pdf
│   ├── Poon2018-UC-LTM\ Unidimensional\ Clustering\ Using\ Latenta.pdf
│   ├── Zhang2017-Latent\ Tree\ Analysis.pdf
│   └── liu-n-ecml14.pdf
├── sample.arff
├── sample.dict-0.csv
├── sample.dict-1.csv
├── sample.dict-2.csv
├── sample.files.txt
├── sample.html
├── sample.nodes.js
├── sample.sparse.txt
├── sample.txt
├── sample.whole_dict-0.csv
├── sample.whole_dict-1.csv
├── sample.whole_dict-2.csv
└── topic_output
    ├── TopicBase.txt
    ├── TopicsTable-Level-1.html
    └── TopicsTable.html

from hlta.

I got your point, you are talking about the sample.html available in the base directory.
My point is that the content of the sample.html is the same as the content of topicstables.html and topicstable-level-1.html,

Sorry for taking your time,
if you think it may be because of my dataset and parameters which created just one level of data, I will try to change them

from hlta.

thank you very much, I will go through that paper also,

I will work on that :)
Many thanks for your time,

from hlta.

I have uploaded a main method for running the computing of topic coherence score. See Readme for more details.

from hlta.