My python implementation of Multilayer Perceptron (MLP) and Long Short Term Memory (LSTM) followed by a feedforward layer. This is a homework for CMU 10-605.
The goal is character level entity classification. Specifically, we build a classifier for article titles in the DBPedia data set. We use the following 5 DBPedia categories:
Person, Place, Organisation, Work, Species.
Please use run.sh to run the code.
The data set released in this repository is a tiny example. Please replace them with your favorate data sets. For the code to work, data must be stored using the same prefix for train, valid and test sets. For example:
data/mydata.train
data/mydata.valid
data/mydata.test
Each data file contains two columns, separated by tab. The first column contains the title, a string without spaces. The second column contains the label name, also a string without spaces. For example:
Lloyd_Stinson Person
Lobogenesis_centrota Species
Loch_of_Craiglush Place
The predicted probability on test data set is stored in a .npy file,
where the output filename can be specified by the --output_file option.