Skip to content

Latest commit

 

History

History
120 lines (87 loc) · 4.19 KB

README.md

File metadata and controls

120 lines (87 loc) · 4.19 KB

Welcome to Sortable Challenge 2016 README

This is Albert Heinle's submission to the challenge as given by the company "Sortable" (http://sortable.com/challenge/).

System requirements

I programmed this submission in Python 3. The version I tested it with is 3.4.3. As additional package(s) that might be installed via pip3, I made use of

  • nltk (http://www.nltk.org/) - Natural language processing for Python. Steps to install::

      $> pip3 install nltk
  $> python3
  Python 3.4.3 (default, Oct 14 2015, 20:28:29) 
  [GCC 4.8.4] on linux
  Type "help", "copyright", "credits" or "license" for more information.
  >>> import nltk
  >>> nltk.download()
  NLTK Downloader
  ---------------------------------------------------------------------------
  d) Download   l) List    u) Update   c) Config   h) Help   q) Quit
  ---------------------------------------------------------------------------
  Downloader> d all

    and wait. Afterwards, exit the downloader and Python.

  • Sphinx (http://www.sphinx-doc.org/en) for the generation of the documentation (optional)

Remark: The d all command downloads the whole nltk functionality, and that might take a moment (depending on your internet connection).

Another Remark: If you have trouble to install nltk, I have written my own primitive tokenizer in the branch without_nltk of this repository. However, this tokenizer is not as accurate, and one might have less sound results compared to the results we would get with nltk.

A Quick Test:

All modules have respective test modules. If you have nosetests3 installed, you can run::

    $> nosetests3 -v

inside the src folder of this project and if everything is OK, your system is properly set up.

How to run:

You need two files to run the given script, namely a file containing listings, and a file containing products. I included the sortable challenge files into the project:

  • test_files/sortable_listings.txt (The listings)
  • test_files/sortable_products.txt (The products)

The script that tries to do the matchings is called run_matching.py in the folder src. In order to run it with the sortable challenge files, go to the folder src and type::

    $> python3 run_matching.py -p ../test_files/sortable_products.txt -l ../test_files/sortable_listings.txt

After that, there will be a folder results generated, with a subfolder named after the timestamp when the run_matching.py script was started. The script generates a couple of threads, each working on a chunk of the listings at a time and writing their results into this folder. In the end, there will be one results.txt file left (and a log file), which contains the results of the computation.

Remark: There are certain parameters that can be set for the project (like number of threads, size of the chunk that each thread is working on, output folder path etc.). These parameters can be set in src/constants.py.

Generating the Documentation

I used Sphinx to auto-generate the documentation of the project. In order to obtain a documentation, go to the subfolder doc and run::

    $> make html

to get a documentation in HTML format. One can find the root documentation file afterwards in doc/_build/html/index.html.

REMARK: Make sure your Sphinx version uses Python3 to interpret the module files. Otherwise, there might be errors when running the above command.

Results:

I ran the script ran on my machine (Intel core i5, ~2.4GHz, 16GB RAM) on both versions that I have (with NLTK and without).

  • With NLTK, it took about 2 minutes total time for the challenge data. The results file produced was about 1.4MB big.
  • Without NLTK, the computation took 30s and created a same size result file as with NLTK.

TODOs:

  • Make script more robust against corrupt files (the challenge files were fortunately not corrupted)
  • Currently, NLTK is not used in its full potential. Introduce language recognition of postings and adapt algorithm accordingly (like information separator words we are currently using, e.g. 'for')