Datasets:

zjt24
/

demon_attack_pi0_validation

tags:
  - tsfile
  - timeseries
  - lerobot
  - atari
  - demon-attack
  - pi0
  - validation
modality: timeseries
size_categories:
  - 10K<n<100K
pretty_name: Demon Attack Pi0 Validation (TsFile)
configs:
  - config_name: default
    data_files:
      - split: validation
        path: data/demon_attack_pi0_validation.tsfile

Demon Attack Pi0 Validation (TsFile)

This dataset contains the numeric time-series data from talha1503/demon_attack_pi0_validation, converted from its LeRobot v2.0 Atari layout to one Apache TsFile. The task asks an agent playing Demon Attack from a single game image to choose one of NOOP, FIRE, RIGHT, LEFT, RIGHTFIRE, or LEFTFIRE, avoid enemy fire, survive, and shoot enemies to maximize score.

Modalities: Time-series in this repository; RGB images in the original repository
Split: validation
Scale: 29 episodes and 51,257 frames/rows
Sampling rate: 30 FPS
Source files: 29 episode Parquet files
Converted files: 1 TsFile
Source revision: 1a423750d2957d33cd2ec4c4931465eb8ab845b6
License: not specified by the source repository

Data File

The complete converted numeric dataset is stored at data/demon_attack_pi0_validation.tsfile. It contains the table demon_attack_pi0_validation. Each source frame maps to one row, while episode and task identifiers form the TsFile TAG dimensions.

TsFile Schema

Column	TsFile role	Type	Description
`Time`	TIME	INT64 timestamp (ms)	`round(timestamp * 1000)` within each episode
`episode_index`	TAG	STRING (source INT64)	Source episode index, `0` through `28`
`task_index`	TAG	STRING (source INT64)	Source task index; `0` for this dataset
`frame_index`	FIELD	INT64	Frame index within the episode
`sample_index`	FIELD	INT64	Source global sample `index`
`done`	FIELD	INT64	Source boolean terminal flag encoded as `0` or `1`
`is_success`	FIELD	INT64	Source boolean success flag encoded as `0` or `1`
`state_0`	FIELD	FLOAT	The one-dimensional source `state` vector
`actions_0` ... `actions_5`	FIELD	FLOAT	The six-dimensional source `actions` vector

Conversion Notes

Source timestamp is represented by the TsFile Time column in milliseconds and is not duplicated as a FIELD.
Source index is renamed to sample_index.
state is flattened to state_0 and actions is flattened to actions_0 through actions_5, using single-precision FLOAT fields.
Source boolean columns done and is_success are stored as INT64 0/1 flags for reliable TsFile readback.
episode_index and task_index are TAG columns, allowing all 29 episodes to remain in one TsFile.
Detailed source-to-TsFile mappings are recorded in meta/info.json.

Images

The source image feature is a 160 x 160 x 3 RGB image embedded in each episode Parquet file. It is intentionally omitted from this numeric TsFile conversion rather than expanded into 76,800 scalar values per frame. The images remain available in the original dataset. The preserved episode, frame, task, and sample indices maintain frame-level alignment with the source rows. The source repository contains no MP4 files.

Reading the TsFile

from tsfile import TsFileReader

reader = TsFileReader("data/demon_attack_pi0_validation.tsfile")
columns = [
    "episode_index",
    "task_index",
    "frame_index",
    "sample_index",
    "done",
    "is_success",
    "state_0",
    "actions_0",
]

with reader.query_table(
    "demon_attack_pi0_validation", columns, batch_size=65536
) as result:
    first_batch = result.read_arrow_batch()
    print(first_batch)

Source and Attribution

The source repository did not provide a dataset card, paper, citation, author list, or license metadata. Its available LeRobot metadata, task description, episode Parquet files, and embedded images are preserved at talha1503/demon_attack_pi0_validation.