Datasets:
Upload README.md with huggingface_hub
Browse files
README.md
CHANGED
|
@@ -1,45 +1,89 @@
|
|
| 1 |
---
|
| 2 |
-
|
| 3 |
-
|
| 4 |
-
|
| 5 |
-
|
| 6 |
-
|
| 7 |
-
|
| 8 |
-
|
| 9 |
-
|
| 10 |
-
|
| 11 |
-
|
| 12 |
-
|
| 13 |
-
|
| 14 |
-
- name: norm_key
|
| 15 |
-
dtype: string
|
| 16 |
-
- name: occurrence
|
| 17 |
-
dtype: int64
|
| 18 |
-
- name: kind
|
| 19 |
-
dtype: string
|
| 20 |
-
- name: dep_fingerprint
|
| 21 |
-
dtype: string
|
| 22 |
-
- name: deps
|
| 23 |
-
list: string
|
| 24 |
-
- name: vtime
|
| 25 |
-
dtype: float64
|
| 26 |
-
- name: provenance
|
| 27 |
-
dtype: string
|
| 28 |
-
- name: payload_json
|
| 29 |
-
dtype: string
|
| 30 |
-
- name: env_id
|
| 31 |
-
dtype: string
|
| 32 |
-
- name: query
|
| 33 |
-
dtype: string
|
| 34 |
-
splits:
|
| 35 |
-
- name: train
|
| 36 |
-
num_bytes: 459200
|
| 37 |
-
num_examples: 1783
|
| 38 |
-
download_size: 109612
|
| 39 |
-
dataset_size: 459200
|
| 40 |
configs:
|
| 41 |
- config_name: default
|
| 42 |
-
data_files:
|
| 43 |
-
- split: train
|
| 44 |
-
path: data/train-*
|
| 45 |
---
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
---
|
| 2 |
+
license: apache-2.0
|
| 3 |
+
task_categories:
|
| 4 |
+
- reinforcement-learning
|
| 5 |
+
tags:
|
| 6 |
+
- groundhog
|
| 7 |
+
- record-replay
|
| 8 |
+
- determinism
|
| 9 |
+
- variance-reduction
|
| 10 |
+
- agent-rl
|
| 11 |
+
- verifiers
|
| 12 |
+
size_categories:
|
| 13 |
+
- n<1K
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 14 |
configs:
|
| 15 |
- config_name: default
|
| 16 |
+
data_files: data/train-*
|
|
|
|
|
|
|
| 17 |
---
|
| 18 |
+
|
| 19 |
+
# Groundhog world tapes -- `groundhog/research-publish-v1`
|
| 20 |
+
|
| 21 |
+
Recorded exogenous randomness for a deliberately non-deterministic agent environment.
|
| 22 |
+
Each row is one intercepted interaction with the outside world: a clock reading, an
|
| 23 |
+
entropy draw, a UUID, an HTTP response, a file mtime, or an environment-variable read.
|
| 24 |
+
|
| 25 |
+
**These tapes replay with no API key, no server and no network.** That is checked before
|
| 26 |
+
publication: the mock world is shut down and every tape is replayed against a closed port.
|
| 27 |
+
|
| 28 |
+
## Contents
|
| 29 |
+
|
| 30 |
+
| | |
|
| 31 |
+
|---|---|
|
| 32 |
+
| worlds (distinct `world_seed`) | 8 |
|
| 33 |
+
| events | 1783 |
|
| 34 |
+
| branches per world | 4 |
|
| 35 |
+
| corpus documents | 30 |
|
| 36 |
+
| live API calls spent recording | 155 |
|
| 37 |
+
| live API calls to replay | **0** |
|
| 38 |
+
| groundhog version | 0.1.0 |
|
| 39 |
+
| seeds | 20240202, 20240203, 20240204, 20240205, 20240206, 20240207, 20240208, 20240209 |
|
| 40 |
+
|
| 41 |
+
## Schema
|
| 42 |
+
|
| 43 |
+
| column | meaning |
|
| 44 |
+
|---|---|
|
| 45 |
+
| `world_seed` | which world instance; every derived exogenous value is keyed by it |
|
| 46 |
+
| `branch` | `root` for the shared fork-point prefix, `root/bN` for a branch |
|
| 47 |
+
| `step` | position in that branch's history |
|
| 48 |
+
| `channel` | `time`, `random`, `uuid`, `http`, `fs`, `env`, `latency` |
|
| 49 |
+
| `norm_key` | normalised request key, e.g. `GET /search?k=3&q=solar` |
|
| 50 |
+
| `occurrence` | how many times this branch had already asked that key |
|
| 51 |
+
| `kind` | `action_independent` (shareable across branches) or `action_dependent` |
|
| 52 |
+
| `dep_fingerprint` | hash of the resource versions *this branch* mutated; `""` means shareable |
|
| 53 |
+
| `deps` | resources whose state can affect the answer |
|
| 54 |
+
| `provenance` | `live`, `tape`, `derived`, `synth`, `refetch`, `approx`, `mutation` |
|
| 55 |
+
| `payload_json` | the recorded value |
|
| 56 |
+
|
| 57 |
+
`kind` and `dep_fingerprint` are the interesting columns. An empty fingerprint means no
|
| 58 |
+
action the branch took can have influenced that answer, so every sibling branch may reuse
|
| 59 |
+
it verbatim -- which is where the variance reduction comes from.
|
| 60 |
+
|
| 61 |
+
## Replay it offline
|
| 62 |
+
|
| 63 |
+
```python
|
| 64 |
+
from datasets import load_dataset
|
| 65 |
+
from groundhog import WorldTape, Mode
|
| 66 |
+
from envs.research_env import ResearchEnv, make_session
|
| 67 |
+
|
| 68 |
+
rows = load_dataset("NagaYu/groundhog-world-tapes", split="train")
|
| 69 |
+
rows = rows.filter(lambda r: r["world_seed"] == rows[0]["world_seed"])
|
| 70 |
+
tape = WorldTape.from_records(rows, "<memory>", branches=[{"branch_id": "root"}])
|
| 71 |
+
|
| 72 |
+
env = ResearchEnv("http://127.0.0.1:9/") # nothing listens there, on purpose
|
| 73 |
+
s = make_session(Mode.GROUNDHOG, tape.world_seed, env, tape=tape, branch_id="root/b0")
|
| 74 |
+
with s.activate():
|
| 75 |
+
result = env.rollout(s, "verify", "solar")
|
| 76 |
+
assert s.stats.live_calls == 0 # no network was touched
|
| 77 |
+
```
|
| 78 |
+
|
| 79 |
+
## Provenance and terms of service
|
| 80 |
+
|
| 81 |
+
Recorded from the bundled mock world in `envs/` -- no real third-party service was
|
| 82 |
+
contacted, and no credentials or personal data are present. Recording a real service with
|
| 83 |
+
the same interceptors is possible and is **the user's responsibility**: check that
|
| 84 |
+
provider's terms first, and do not redistribute content you have no right to.
|
| 85 |
+
|
| 86 |
+
## Citation
|
| 87 |
+
|
| 88 |
+
Part of the [Groundhog](https://github.com/NagaYu/groundhog) research prototype:
|
| 89 |
+
*share the world, resample the consequences.*
|