Datasets:
File size: 3,057 Bytes
d846fa6 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 | {
"meta_info": {
"case_id": "ISRUC_05.edf",
"bench_subset": "NeuroBench-Core",
"difficult": 1.5,
"original_dataset": "ISRUC"
},
"agent_input": {
"data_path": "data/core/ISRUC_05.edf",
"instruction": "Please select standard EEG channels. In the delta, theta, alpha, and beta bands, compute channel-wise relative power and identify the top-3 channels for each band. Return the set of channels that appear in the top-3 list for at least two bands. Use Welch PSD estimation. Define relative power as the integrated power in the target frequency band divided by the integrated total power from 0.5 to 45 Hz."
},
"eval_config": {
"parser_prompt": "### ROLE\nYou are a precise JSON Extraction Engine for neurophysiological data. Your sole task is to convert natural language reports into structured JSON data.\n\n### TASK\nExtract the final set of channels that appear in the top-3 list for at least two bands.\n\n### STRICT CONSTRAINTS (MANDATORY)\n1. Output ONLY a valid JSON object.\n2. DO NOT include Markdown code blocks.\n3. DO NOT include any conversational text or explanations.\n4. The key must be EXACTLY \"multi_band_top_channels\".\n5. Extract the explicitly reported FINAL cross-band set, regardless of its source field name. A final field named channels, channels_in_at_least_two_bands, channels_in_top3_at_least_2_bands, or equivalent is valid.\n6. Do not return any single band's top-3 list and do not merge all per-band intermediate lists. The value must be an array of strings.\n7. Normalize EACH selected channel independently: trim whitespace, remove one leading acquisition prefix EEG followed by a space, hyphen, or underscore, then uppercase the label.\n8. Remove one terminal reference suffix if present: -A1, -A2, -LE, -REF, -M1, -M2, or -AVG.\n9. Convert legacy aliases after removing the prefix and suffix: T3->T7, T4->T8, T5->P7, T6->P8.\n10. Each output label must contain only uppercase A-Z letters and digits. Example: {\"channels\":[\"O1-A2\",\"EEG T4-LE\"]} -> [\"O1\",\"T8\"].\n11. Preserve the final set order as reported. If the report explicitly gives an empty final set, return an empty array.\n12. If no explicit final cross-band set can be identified, return {\"multi_band_top_channels\": null}; do not guess from intermediate lists.\n\n### OUTPUT TEMPLATE\n{\"multi_band_top_channels\": [\"CH1\", \"CH2\"]}",
"metrics": [
{
"metric_id": "multi_band_top_channel_set",
"type": "set_match_check",
"target_key": "multi_band_top_channels",
"weight": 100,
"params": {
"gt_value": [
"C3",
"C4",
"F4",
"O2"
],
"match_mode": "per_hit_fraction",
"top_k": 4,
"score_per_hit": 0.25
}
}
]
}
} |