dquarel commited on
Commit
06eeff8
·
1 Parent(s): f1ce2ef

Document missing runs in README and add train_missing.yaml

Browse files
Files changed (2) hide show
  1. README.md +13 -1
  2. train_missing.yaml +41 -0
README.md CHANGED
@@ -54,13 +54,25 @@ wandb_project=jaxgmg2_3phase_fast_tight
54
 
55
  Checkpoints are named `al_{alpha}_g_{discount_rate}_seed_{seed}_pa_1`.
56
 
 
 
 
 
 
 
 
 
 
 
57
  ## Reproduced with
58
 
59
- See `train.yaml` (initial sweep) and `train_extended.yaml` (extended sweep) in this repository. Run with:
 
60
 
61
  ```bash
62
  timaeus run train.yaml
63
  timaeus run train_extended.yaml
 
64
  ```
65
 
66
  from the [timaeus monorepo](https://github.com/timaeus-research/timaeus).
 
54
 
55
  Checkpoints are named `al_{alpha}_g_{discount_rate}_seed_{seed}_pa_1`.
56
 
57
+ ## Missing runs
58
+
59
+ 5 runs from the initial sweep failed and are not present:
60
+
61
+ - `al_0.0_g_0.97_seed_100_pa_1`
62
+ - `al_0.0_g_0.99_seed_100_pa_1`
63
+ - `al_0.45_g_0.97_seed_103_pa_1`
64
+ - `al_0.45_g_0.98_seed_100_pa_1`
65
+ - `al_0.4_g_0.99_seed_104_pa_1`
66
+
67
  ## Reproduced with
68
 
69
+ See `train.yaml` (initial sweep), `train_extended.yaml` (extended sweep), and
70
+ `train_missing.yaml` (5 failed runs from initial sweep) in this repository. Run with:
71
 
72
  ```bash
73
  timaeus run train.yaml
74
  timaeus run train_extended.yaml
75
+ timaeus run train_missing.yaml
76
  ```
77
 
78
  from the [timaeus monorepo](https://github.com/timaeus-research/timaeus).
train_missing.yaml ADDED
@@ -0,0 +1,41 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ parameters:
2
+ project_name: jaxgmg2_3phase_fast_tight
3
+ action: rl
4
+ rl_action: train
5
+
6
+ lr: 5e-5
7
+ cheese_loc: any
8
+ env_layout: open
9
+ mask_type: first_episode
10
+ use_prev_action: false
11
+
12
+ num_total_env_steps: 10_000_000_000
13
+ num_levels: 9600
14
+ grad_acc_per_chunk: 4
15
+ num_rollout_steps: 64
16
+
17
+ ckpt_dir: jaxgmg2_3phase_fast_tight
18
+ eval_schedule: "0:1,250:2,500:5,2000:10"
19
+ f_str_ckpt: "al_{alpha}_g_{discount_rate}_seed_{seed}_pa_1"
20
+
21
+ use_wandb: true
22
+ use_hf: true
23
+ wandb_project: jaxgmg2_3phase_fast_tight
24
+ ntfy: david_jaxgmg
25
+
26
+ sweep:
27
+ - - alpha: 0.0
28
+ discount_rate: 0.97
29
+ seed: 100
30
+ - alpha: 0.0
31
+ discount_rate: 0.99
32
+ seed: 100
33
+ - alpha: 0.45
34
+ discount_rate: 0.97
35
+ seed: 103
36
+ - alpha: 0.45
37
+ discount_rate: 0.98
38
+ seed: 100
39
+ - alpha: 0.4
40
+ discount_rate: 0.99
41
+ seed: 104