This interactive visualization displays information from the LDA topic models generated using the InPhO VSM module. 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 http://inphodata.cogs.indiana.edu.
The color bands within each article's row show the topic distribution within that article, and the relative sizes of each band indicates the weight of that topic in the article. 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.
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
-
Go to the Anaconda website and choose the Python 2.7 Distribution. (Windows users: during installation, during "Advanced Options" choose "Add Anaconda to my PATH environment variable" and don't worry about the warning.)
-
Open a Terminal (Mac and Linux) or PowerShell (Windows).
-
Run
pip install --pre topicexplorer -i https://inpho.cogs.indiana.edu/pypi/
.Note:
--pre
has two-
characters. When the1.0
release happens,--pre
will no longer be necessary. -
Test installation by typing
topicexplorer -h
to print usage instructions.
- Set up Git
- Install the Anaconda Python 2.7 Distribution.
- Open a terminal and run
pip install --src . -e git+https://github.com/inpho/topic-explorer#egg=topicexplorer
- Test installation by typing
topicexplorer -h
to print usage instructions.
-
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.
-
Train LDA models using the on-screen instructions:
topicexplorer train CONFIG
-
Launch the topic explorer:
topicexplorer launch CONFIG
-
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.
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.
We highly recommend using the Anaconda Python 2.7 Distribution. Straightforward instructions are provided above for Anaconda Python 2.7 for both end users and developers. Both of these installs are officially supported.
Below we offer guidance for installing side-by-side with an Anaconda Python 3.5 install or for installing it without Anaconda, with notes on dependencies.
The InPhO Topic Explorer is only compatible with Python 2.7. However, Anaconda for Python 3.5 makes it easy to set up a side-by-side install of Python 2.7 so you can use both Python 3.5 and Python 2.7.
- Install the Anaconda Python 3.5 Distribution.
- Open a terminal and run
conda create -n py27 python=2.7 anaconda
. This will create a Python 2.7 Anaconda environment. - Run
source activate py27
to activate the Python 2.7 bindings. You should see(py27)
before your prompt. - Use either the Default or Developer install instructions, skipping the step to install Anaconda Python 2.7.
- Run
source deactivate
to deactivate Python 2.7 bindings and reactivate Python 3.5 bindings. Note that thetopicexplorer
command will only work when the Python 2.7 bindings are activated.
-
Miniconda
- If using Miniconda (a small version of Anaconda), the necessary packages are:
conda install numpy scipy nltk matplotplib ipython networkx
- If using Miniconda (a small version of Anaconda), the necessary packages are:
-
Debian/Ubuntu
-
sudo apt-get-install build-essential python-dev python-pip python-numpy python-matplotlib python-scipy python-ipython
-
-
Windows
-
Install the Python packages below:
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.
- Install apache2 with mod_wsgi:
apt-get install apache2 libapache2-mod-wsgi
- Create
/etc/apache2/sites-available/topicexplorer.conf
:
<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=/home/jaimie/anaconda2/lib/python2.7/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>
Credit to Digital Ocean for a very helpful guide.
To enable caching:
- 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>
- Enable caching modules:
sudo a2enmod cache && sudo a2enmod cache_disk
- Restart apache:
sudo apache2ctl restart
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>
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://inphodata.cogs.indiana.edu/
A BibTeX file is included in the repository for easier attribution.
The InPhO Topic Explorer is maintained by Jaimie Murdock:
- E-mail: jammurdo@indiana.edu
- Twitter: @JaimieMurdock
- GitHub: @JaimieMurdock
- Homepage: http://jamram.net/
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.