rgalhama / retro_adjs

Code for Retrodiction as Delayed Recurrence

Ref to paper (ToDo - complete)

Retrodiction as Delayed Recurrence: the Case of Adjectives in Italian and English

Requirements

This code has only been tested in Linux Mint 19+, with the Python library versions specified in environment.yml, and R version 4.0.4.

Getting started

The quick recipe:

• Install Miniconda3 (or Conda if preferred):

https://docs.conda.io/en/latest/miniconda.html

• Create an environment using the provided environment.yml file:

conda env create -f environment.yml 

• You may need to install PyTorch and the Spacy models separately:

conda install --name learningadjs pytorch==1.5.0 torchvision==0.6.0 -c pytorch

Alternatively, use pip from within your activated (virtual) Python environment:

cat requirements.txt | xargs -n 1 pip install 
pip install torch==1.5.0+cu101 torchvision==0.6.0+cu101 -f https://download.pytorch.org/whl/torch_stable.html

The reported simulations used Spacy v. 2.0.12.

conda install -c conda-forge spacy
python -m spacy download en_core_news_sm
python -m spacy download it_core_news_sm

The following packages are required for AoA analyses:

sudo apt-get -y install r-cran-rmysql libcurl4-gnutls-dev libxml2-dev libssl-dev libmysql++-dev gfortran liblapack-dev liblapack3 libopenblas-base  libopenblas-dev 

As well as:

install.packages("magrittr") 
install.packages("dplyr")    
install.packages('devtools', repos='http://cran.rstudio.com/')
install.packages("stringr") #required for wordbankr
install.packages("wordbankr")
install.packages("optparse")

For using R from Bash scripts and command line:

sudo apt-get install littler

Reproducing Results

To reproduce the analyses in the paper, :

1. Data Preparation
2. Analyses TPs
3. Analyses AoA
4. Train and test RNNs (see src/scripts)
5. Analyze predictions of RNNs