We provide an extension neural and probabilistic models of recursive models for tree-structure data based on tensor theory.
List of publication based on this repository:
-
"Generalising Recursive Neural Models by Tensor Decomposition",
Daniele Castellana, Davide Bacciu,
International Joint Conference on Neural Networks (IJCNN), Glasgow, UK, 2020 -
"Tensor Decompositions in Recursive Neural Networks for Tree-Structured Data",
Daniele Castellana, Davide Bacciu,
European Symposium on Artificial Neural Networks, Computational Intelligence and Machine Learning, Bruges, Belgium, 2020 -
"Learning from Non-Binary Constituency Trees via Tensor Decomposition",
Daniele Castellana, Davide Bacciu,
Proceedings of the 28th International Conference on Computational Linguistics, Barcelona, Spain, 2020
The code is structured as follows:
data
contains the BoolSent datasetmodels
contains the implementation of the recursive models (both probabilistic and neural);preprocessing
contains the code to preprocess the datasets;tasks
contains all the files to execute the experiments; there is a readme files for each task;tree_utils
contains utils code.
Also, two other repositories are used:
exputils
provides the utils to run an experiment, train a model, parse configuration, etc..thlogprob
provides a minimal library to handle distribution using pytorch.
See the readme for more information.
-
download the raw data (see next section);
-
in NLP tasks, sentences should be parsed running the command:
python tasks/task_name/parse_raw_data raw_data_folder output_folder
-
Run the preprocessor using the command:
python preprocess.py --config-file preproc_config_file
,
wherepreproc_config_file
can be found in the foldertasks/task_name/config_files
-
Run the experiment using the command:
python run.py --config-file run_config_file
,
whererun_config_file
can be found in the foldertasks/task_name/config_files
.