reorderdata/reordering_model

Deep Reordering Model

The extensive and diverse matrices in ReorderData offer valuable supervision for training a deep reordering model. By treating the matrices with index swaps as negative samples and their ground-truth matrices as positive samples, we build a deep model for matrix reordering.

Quick start

1. Download data

Download the ReorderData test set from here, the unified scoring model from here, and the source code from here.

2. Setup environment

# install from requirements.txt
pip3 install -r requirements.txt

3. Run

python test.py \
    --data_path <path> \ # the path to swap_dic.npz file
    --model_path <path> \ # the path to the deep reordering model checkpoint
    --scorer_path <path> \ # the name to the convext-tiny unified scoring model checkpoint
    --pattern_type <pattern_type> \ # one of block, offblock, star, and band
    --continuous_eval \ # for continuous matrices only