Layout_Generator

Usage

Build

make -B

Generate Cases

Edit training set specs in case_gen.py:

# (test_num, width, height, layers, obs_num, min_obs_size, max_obs_size, net_num, pin_num)
levels = [
    (2000, s, s, l, int(s * 0.50), min_obs_size, max_obs_size, 1, 4),
    (800, s, s, l, int(s * 0.50), min_obs_size, max_obs_size, 15, 5),
    (160, s, s, l, int(s * 0.25), min_obs_size, max_obs_size, 75, 5),
    (80, s, s, l, int(s * 0.10), min_obs_size, max_obs_size, 150, 5),
    (40, s, s, l, int(s * 0.05), min_obs_size, max_obs_size, 300, 5),
]  # level_0 must be single net and other should not```

where:

• test_num: number of cases in that level (2000 cases in level_0)
• net_num: number of nets in any case in that level (any case in level_0 has 1 net)
• pin_num: number of pins in each net in any case in that level (any net in level_0 has 4 pins)

Similar for eval & test set:

gen_eval(
    args.size,
    args.layer,
    (40, s, s, l, int(s * 0.50), min_obs_size, max_obs_size, 1, 5),
)
gen_test(
    args.size,
    args.layer,
    (40, s, s, l, int(s * 0.05), min_obs_size, max_obs_size, 300, 5),
)

Then execute this command to generate data set:

python case_gen.py --main MAIN --size SIZE --layer LAYER

default:

• MAIN = main
• SIZE = 500
• LAYER = 3

Structure of Datasets

Here is an example of a training set

train_500x500x3
├── level_0
│   ├── id_10000.txt
│   ...
├── level_1
├── level_2
├── level_3
├── level_4
├── raw  # raw info, does not required by RL env
│   ├── level_0
│   │   ├── 0.png  # first case has visualized layout
│   │   ├── 0.txt
│   │   ├── 1.txt
│   │   ...
│   ├── level_1
│   ├── level_2
│   ├── level_3
│   └── level_4
├── config.pickle  # file required by RL env
└── config.txt  # file required by RL env