core

Generating Subjective Content Descriptions

We compare two approaches, the SCD-matrix based MPSCD and iSCD versus

Use the code

  • core.corpus contains the classes to access the corpora and annoate them with SCDs
  • core.download contains the code to download all needed files, like corpora, models and other files
  • core.evaluation contains the code to evaluate the different settings (wrapper around core.model)
  • core.experiments contains the code to run different experiments (entrypoint files!)
  • core.model contains the models and all code related to each individual model
  • core.utils contains helper classes, functions and global constants

Run the code

We assume the docker container was started and a shell is opened inside.

  1. Download needed files with core.download by running /home/user/core/setup.py
  2. Run a sample setup (consiting of fine tuning/ training models and evaluating the performance) via /home/user/core/main.py
  3. Results will be written as JSON into /home/user/core/results/
  4. Run real evaluation and combine multiple results in /home/user/core/results/ via /home/user/core/experiments/*.py

To create own runs, take a look at core.model.exec.Executor and the source of core.main.

View Source 
"""
# Generating Subjective Content Descriptions

We compare two approaches, the SCD-matrix based MPSCD and 
iSCD versus 

## Use the code
- `core.corpus` contains the classes to access the corpora and annoate them with SCDs
- `core.download` contains the code to download all needed files, like corpora, models and other files
- `core.evaluation` contains the code to evaluate the different settings (wrapper around `core.model`)
- `core.experiments` contains the code to run different experiments (entrypoint files!)
- `core.model` contains the models and all code related to each individual model
- `core.utils` contains helper classes, functions and global constants

## Run the code
> We assume the docker container was started and a shell is opened inside. 

1. Download needed files with `core.download` by running ``/home/user/core/setup.py``
2. Run a sample setup (consiting of fine tuning/ training models and evaluating the performance) via ``/home/user/core/main.py``
3. Results will be written as JSON into ``/home/user/core/results/``
4. Run real evaluation and combine multiple results in ``/home/user/core/results/`` via ``/home/user/core/experiments/*.py``

> To create own runs, take a look at `core.model.exec.Executor` and the source of `core.main`.
"""