Checkpoint 2

Maps generation

We adopted TMP (Tiled Motion Planning) dataset generation script add modified hyperparams as showed below:

python generate_spp_instances.py 
    --input-path "data/mpd/original/*" 
    --output-path data/mpd/ 
    --maze-size 128 
    --mechanism moore 
    --edge-ratio 0.25 
    --tile-size=2 
    --train-size 800 
    --valid-size 100 
    --test-size 100

Examples:

Tasks generation

For every map we generate 10 start/goal nodes with hardness > 1.05

Task generation script

python task_generation.py --maps_path <path_to_maps>

Model training

FS+PPM model

Get focal values:

python get_focals.py
    --filename ./data/1k_128x2_v2/

Training:

python lib.TransPath.train.py 
    --mode f
    --batch 16

View FS+PPM model training report:
https://api.wandb.ai/links/blain/4679lk9x

Model evaluation

Detailed evaluation of FS+PPM model:
https://github.com/LogicZMaksimka/TransPath-Project/blob/master/visualise_metrics.ipynb