yashkgp / active_zero_shot_learning

Code for the paper titled "Distributed representation of tags for Active Zero Shot learning"

Download Datset

Dataset Link - https://archive.org/download/stackexchange

Data files to download:

• Dba - dba.stackexchange.com.7z
• Unix - unix.stackexchange.com.7z

Extract only the files Tags.xml and Posts.xml .

Put the two files of each of the 2 subsystems in the following data folders - data/dbaData and data/unixData respectively.

PreProcessing -

Run python prepare_data.py

to perform pre-processing over the data files and generate the required files for the experiment. You will need to change the parameter DATA_DIR in prepare_data.py to either data/dbaData or data/unixData depending the subsystem you are working wi th. The output is generated in data/dbaData/output or data/unixData/output respectively.

Experiment -

In order to run the experiment you need to run the command - python main.py --data_dir=<data_dir> --start_seen=<start_seen> --end_seen=<end_seen> --plot_file=<plot_file> --measure= <centrality_measure> where,

<data_dir> - is the directory where data is present. It will be data/dbaData/output or data/unixData/output depending on the subsystem.

<start_seen> - Number of Seen Classes starting range

<end_seen> - Number of Seen Classes ending range

<plot_file> - name of the output plot (Precision @5 vs Number Seen Classes) that will be generated.

<centrality_measure> - name of the centrality measure to use

The code reads the data and then creates a similarity_matrix using the boltzman machine. If the file similarity_matrix.npy is already present in data_dir, it skips its recomputation, if it is not present, it trains the similarity_matrix again and saves in the data_dir folder. It then runs the Active Zero Shot Learning Algorithm and gets the Precision @ 5 scores and produces the final plot and png file.

Requirements -

Python 3 (3.6)

Pickle

Numpy

Scipy

xml

Beautiful Soup

json

matplotlib