Skip to content

inpho/topic-explorer

Repository files navigation

InPhO Topic Explorer and Hypershelf

Join the chat at https://gitter.im/inpho/topic-explorer Travis AppVeyor Windows build GitHub license PyPI

The InPhO Topic Explorer provides an integrated system for text modeling making it simple to go from a set of documents to an interactive visualization of LDA topic models generated using the InPhO VSM module. More advanced analysis is made possible by a built-in pipeline to Jupyter (iPython) notebooks.

Live demos trained on the Stanford Encyclopedia of Philosophy, a selection of books from the HathiTrust Digital Library, and the original LDA training set of Associated Press articles are available at https://www.hypershelf.org.

The Hypershelf provides an interactive, document-centered visualization of the topic models. Each document is represented by a multi-colored, horizontal bar representing the overall topic distribution for the document, with different colors representing the topics. The relative lengths of each segment of the bar indicates the weight of the topic in that document. The total width of each row indicates similarity to the focal topic or document, measured by the quantity sim(doc) = 1 – JSD(doc, focus entity), where JSD is the Jensen-Shannon distance between the word probability distributions of each item. Each topic's label and color is arbitrarily assigned, but is consistent across articles in the browser.

Display options include topic normalization, alphabetical sort and topic sort. By normalizing topics, the combined width of each bar expands so that topic weights per document can be compared. By clicking a topic, the documents will reorder acoording to that topic's weight and topic bars will reorder according to the topic weights in the highest weighted document. When a topic is selected, clicking "Top Documents for [Topic]" will take you to a new page showing the most similar documents to that topic's word distribution. The original sort order can be restored with the "Reset Topic Sort" button.

Installation

There are two types of install: Default and Developer. Currently we are debugging the Windows Default installation. Before installing, Mac users should install XCode from Apple and Windows users should install a Python compiler such as Microsoft Visual C__ Compiler

Note: If you attempt to install this while connected to the Internet by way of a restricted network (such as eduroam) you may experience difficulty. Our pypi server requires port 8080 to be accessible, and this is sometimes blocked on public networks.

Default Install (Windows users should use Developer Install until further notice)

  1. Go to the Anaconda website and choose the Python 3.7 Distribution. (Windows users: during installation, during "Advanced Options" choose "Add Anaconda to my PATH environment variable" and don't worry about the warning.)

  2. Open a Terminal (Mac and Linux) or PowerShell (Windows).

  3. Run pip install --pre topicexplorer.

    Note: --pre has two - characters. When the 1.0 release happens, --pre will no longer be necessary.

  4. Test installation by typing topicexplorer -h to print usage instructions.

Developer Install

  1. Set up Git
  2. Install the Anaconda Python 3.7 Distribution.
  3. Open a terminal and run pip install --src . -e git+https://github.com/inpho/topic-explorer#egg=topicexplorer
  4. Test installation by typing topicexplorer -h to print usage instructions.

Usage

Workflow

  1. Initialize the Topic Explorer on a file, folder of text files, or folder of folders:

    topicexplorer init PATH
    

    This will generate a configuration file called CONFIG.

  2. Train LDA models using the on-screen instructions:

    topicexplorer train CONFIG
    
  3. Launch the topic explorer:

    topicexplorer launch CONFIG
    
  4. Press Ctrl+C to quit all servers.

See the sample configuration files in the config directory for examples of how to extend the topic explorer.

Bug Reports

Please report issues on the issue tracker or contact Jaimie directly (contact info at bottom of README).

In your report, please include the error message, the command you ran, your operating system, and the output of the command topicexplorer version. This will ensure that we can quickly diagnose your issue.

Note: When using a developer install topicexplorer version will print in the following format: 1.0b39-1-g7c834bf-dirty.

  • The first part is the most recent release tag. (1.0b39)
  • The second part is the number of commits since the tag. (1)
  • The next is the hash of the most recent commit. (g7c834bf)
  • The optional -dirty flag indicates that the local repository has uncommitted changes.

Alternate Installs

We highly recommend using the Anaconda Python 3.7 Distribution using the straightforward instructions provided above. While we continue to support Python 2.7 installations for the time being, this will not continue indefinitely.

Below we offer guidance for installing the Python 2.7 version alongside Python 3 using Anaconda, and for installing without Anaconda, with notes on dependencies.

Python 2 alongside Python 3

The InPhO Topic Explorer is compatible with Python 2.7. Anaconda for Python 3.6 makes it possible to set up a side-by-side install of Python 2.7 so you can use both Python 3.7 and Python 2.7.

  1. Install the Anaconda Python 3.7 Distribution.
  2. Open a terminal and run conda create -n py27 python=2.7 anaconda. This will create a Python 2.7 Anaconda environment.
  3. Run source activate py27 to activate the Python 2.7 bindings. You should see (py27) before your prompt.
  4. Use either the Default or Developer install instructions, skipping the step to install Anaconda Python 2.7.
  5. Run source deactivate to deactivate Python 2.7 bindings and reactivate Python 3.7 bindings. Note that the topicexplorer command will only work when the Python 2.7 bindings are activated.

Non-Anaconda Install

  • Miniconda

    1. If using Miniconda (a small version of Anaconda), the necessary packages are: conda install numpy scipy scikit-learn nltk matplotplib ipython unidecode wget decorator chardet ujson requests notebook
  • Debian/Ubuntu

    1. sudo apt-get-install build-essential python-dev python-pip python-numpy python-matplotlib python-scipy python-ipython

    2. IPython Notebooks

  • Windows

    1. Install Microsoft Visual C++ Compiler for Python 2.7

    2. Install the Python packages below:

Deployment

The Topic Explorer includes a simple HTTP server through the Bottle Web Framework. This simple server is what the topicexplorer launch and topicexplorer serve commands use.

If you wish to host a high-traffic version of InPhO-TE, we recommend using the Apache Web Server and the mod_wsgi module. This infrastructure powers the demos at [http://inphodata.cogs.indiana.edu/] and is readily available on cloud compute platforms. Details for that configuration are below.

If you have deployed InPhO-TE using another infrastructure, please submit a pull request with the documentation. The function topicexplorer.server.create_app will make a WSGI-compliant application for use with any WSGI-compliant server.

mod_wsgi

  1. Install apache2 with mod_wsgi: apt-get install apache2 libapache2-mod-wsgi
  2. Create /etc/apache2/sites-available/topicexplorer.conf [you will need to fill in your local paths where specified]:
    <VirtualHost *:80>
    	ServerName localhost
    	ServerAdmin admin@localhost
    	
    	ErrorLog /var/www/topicexplorer/log/error.log
    	CustomLog /var/www/topicexplorer/log/access.log combined
    
    	WSGIDaemonProcess topicexplorer user=www-data group=www-data \
    	  python-path=[fill in; e.g. /home/jaimie/anaconda3/lib/python3.7/site-packages/]
    	WSGIScriptAlias /ap /var/www/topicexplorer/app.wsgi
    	
    	<Directory /var/www/topicexplorer>
    		WSGIProcessGroup topicexplorer
    		WSGIApplicationGroup %{GLOBAL}
    		Options All
    		AllowOverride All
    		Require all granted
    	</Directory>
    </VirtualHost>
    
  3. Create the application directory: sudo mkdir -p /var/www/topicexplorer/
  4. Create the log directory: sudo mkdir -p /var/www/topicexplorer/log
  5. Create the www directory: sudo mkdir -p /var/www/topicexplorer/www
  6. Create the config directory: sudo mkdir -p /var/www/topicexplorer/config
  7. Add a symlink from /var/www/topicexplorer/ to your local installation of app.wsgi. For example: ln -s /home/jaimie/workspace/topic-explorer/app.wsgi /var/www/topicexplorer/app.wsgi
  8. Add a symlink to your .ini files in the /var/www/topicexplorer/config directory. For example, if working with the AP corpus trained in my home directory: ln -s /home/jaimie/ap.ini /var/www/topicexplorer/config/ap.ini
  9. Enable the site: sudo a2ensite topicexplorer
  10. Restart apache: sudo apache2ctl restart
  11. Test the site at [http://localhost/]. If an HTTP 500 Internal Server Error is returned, check the log in /var/www/topicexplorer/log/error.log.

Caching

Credit to Digital Ocean for a very helpful guide.

To enable caching:

  1. Modify /etc/apache2/sites-available/topicexplorer.conf to add Cache directives before the closing </VirtualHost> tag:
<VirtualHost *:80>
    # ...

	CacheQuickHandler off

	CacheLock on
	CacheLockPath /tmp/mod_cache-lock
	CacheLockMaxAge 5

	CacheIgnoreHeaders Set-Cookie

	<Location />
		CacheEnable disk
		CacheHeader on

		CacheDefaultExpire 600
		CacheMaxExpire 86400
		CacheLastModifiedFactor 0.5
	</Location>
</VirtualHost>
  1. Enable caching modules: sudo a2enmod cache && sudo a2enmod cache_disk
  2. Restart apache: sudo apache2ctl restart

Server Status

The following snippet may help debug issues, enabling you to visit [http://localhost/server-status] to find more details. Remove these lines for production environments.

<VirtualHost *:80>
    # ...

	<Location /server-status>
		SetHandler server-status
		
		Order Deny,Allow
		Allow from all
	</Location>
</VirtualHost>

Licensing and Attribution

The project is released under an Open-Source Initiative-approved MIT License.

The InPhO Topic Explorer may be cited as:

  • Jaimie Murdock and Colin Allen. (2015) Visualization Techniques for Topic Model Checking in Proceedings of the 29th AAAI Conference on Artificial Intelligence (AAAI-15). Austin, Texas, USA, January 25-29, 2015. http://inphoproject.org/papers/

A BibTeX file is included in the repository for easier attribution.

Collaboration and Maintenance

The InPhO Topic Explorer is maintained by Jaimie Murdock:

Please report issues on the issue tracker or contact Jaimie directly.

We are open to collaboration! If there's a feature you'd like to see implemented, please contact us and we can lend advice and technical assistance.