Document missing control runs, add control_train_missing.yaml
Browse files- README.md +4 -0
- control_train_missing.yaml +42 -0
README.md
CHANGED
|
@@ -32,6 +32,9 @@ env_rule = "ban:c[10,11], m[11,:]; c[11,10], m[:,11]"
|
|
| 32 |
|
| 33 |
See [`control_train.yaml`](./control_train.yaml).
|
| 34 |
|
|
|
|
|
|
|
|
|
|
| 35 |
## Shared Hyperparams
|
| 36 |
|
| 37 |
```
|
|
@@ -60,6 +63,7 @@ use_hf=True
|
|
| 60 |
```bash
|
| 61 |
timaeus run antenna_train.yaml
|
| 62 |
timaeus run control_train.yaml
|
|
|
|
| 63 |
```
|
| 64 |
|
| 65 |
from the [timaeus monorepo](https://github.com/timaeus-research/timaeus).
|
|
|
|
| 32 |
|
| 33 |
See [`control_train.yaml`](./control_train.yaml).
|
| 34 |
|
| 35 |
+
Missing run_ids: 84, 87, 90, 92, 93, 95, 101, 102, 106, 109.
|
| 36 |
+
To fill these, see [`control_train_missing.yaml`](./control_train_missing.yaml).
|
| 37 |
+
|
| 38 |
## Shared Hyperparams
|
| 39 |
|
| 40 |
```
|
|
|
|
| 63 |
```bash
|
| 64 |
timaeus run antenna_train.yaml
|
| 65 |
timaeus run control_train.yaml
|
| 66 |
+
timaeus run control_train_missing.yaml
|
| 67 |
```
|
| 68 |
|
| 69 |
from the [timaeus monorepo](https://github.com/timaeus-research/timaeus).
|
control_train_missing.yaml
ADDED
|
@@ -0,0 +1,42 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
parameters:
|
| 2 |
+
project_name: jaxgmg2_antennaban
|
| 3 |
+
action: rl
|
| 4 |
+
rl_action: train
|
| 5 |
+
|
| 6 |
+
lr: 5e-5
|
| 7 |
+
alpha: 0.6
|
| 8 |
+
discount_rate: 0.98
|
| 9 |
+
cheese_loc: any
|
| 10 |
+
env_layout: open
|
| 11 |
+
env_rule: "ban:c[10,11], m[11,:]; c[11,10], m[:,11]"
|
| 12 |
+
mask_type: first_episode
|
| 13 |
+
use_prev_action: false
|
| 14 |
+
|
| 15 |
+
num_total_env_steps: 10_000_000_000
|
| 16 |
+
num_levels: 9600
|
| 17 |
+
grad_acc_per_chunk: 5
|
| 18 |
+
num_rollout_steps: 64
|
| 19 |
+
|
| 20 |
+
seed_formula: "{int(discount_rate*100):02d}{int(alpha*10):02d}{run_id:03d}"
|
| 21 |
+
ckpt_dir: jaxgmg2_antennaban2
|
| 22 |
+
f_str_ckpt: "control_al_{alpha}_g_{discount_rate}_id_{run_id}_seed_{seed}"
|
| 23 |
+
eval_schedule: "0:1,250:2,500:5,2000:10"
|
| 24 |
+
log_optimizer_state: true
|
| 25 |
+
|
| 26 |
+
wandb_project: jaxgmg2_antennaban
|
| 27 |
+
use_wandb: true
|
| 28 |
+
use_hf: true
|
| 29 |
+
no_tqdm: true
|
| 30 |
+
ntfy: david_jaxgmg
|
| 31 |
+
|
| 32 |
+
sweep:
|
| 33 |
+
- - run_id: 84
|
| 34 |
+
- run_id: 87
|
| 35 |
+
- run_id: 90
|
| 36 |
+
- run_id: 92
|
| 37 |
+
- run_id: 93
|
| 38 |
+
- run_id: 95
|
| 39 |
+
- run_id: 101
|
| 40 |
+
- run_id: 102
|
| 41 |
+
- run_id: 106
|
| 42 |
+
- run_id: 109
|