CompatibilityModeling

Setup

For the setup of this repository simply type:

make

This will

• set up a virtual environment for this repository,
• install all necessary project dependencies.

The lowest python version you can use is 3.7 (because the repository includes data classes).
It is recommended to use Python 3.8.

Clean and Re-install

To reset the repository to its inital state, type:

make dist-clean

This will remove the virtual environment and all dependencies.
With the make command you can re-install them.

To remove temporary files like .pyc or .pyo files, type:

make clean

Packaging Info

The directory src/ is a package, which is installed when you run the make command.
Be careful to run it with the Python module syntax that you can see in the section Running below.

Running

To extract all ambiguous PPs with their possible heads, type:

python3 -m src.extract_ambiguous_pp your_input.conll your_output.tsv

Todo: The explanation for extract_pps must go here.

Testing

To run the tests, type:

python -m pytest tests/test_extract_ambiguous_pp.py

Stored Models

3 of the different trained models have been stored:

1. Logistic Regression
2. NeuralCandidateScoringModel
3. NeuralCandidateScoringModel with Averaged nouns and lemma inputs

The models are stored in the trained-models directory.

The typical pytorch state_dict saving has been used.
To load the model for inference, use:

from src.pp_head_selection.models import NeuralCandidateScoringModel
model = NeuralCandidateScoringModel(input_dim=1076, output_dim=2)
model.load_state_dict(torch.load("trained-models/NeuralCandidateScoringModel"))

# OR:
from src.pp_head_selection.models import NeuralCandidateScoringModel
model = NeuralCandidateScoringModel(input_dim=1076, output_dim=2)
model.load_state_dict(torch.load("trained-models/NeuralCandidateScoringModel-averaged-nouns"))

# OR: 
from src.pp_head_selection.models import LogisticRegression
model = LogisticRegression(input_dim=1076, output_dim=2)
model.load_state_dict(torch.load("trained-models/LogisticRegression"))


# To inspect the model, use:
model.eval()

You must provide the input_dim and output_dim to the models. The input and output dimensions are 1076 and 2 for all models. However, note that for other feature selection processes, the input dimension might change.

You can find the training metrics and metadata for the threem models in train-results/logs.txt.
For the stored models a category Stored Model Path is among the logged information.