MANIFEST.json

{
  "name": "SimplexTasks-12",
  "slug": "simplextasks-12",
  "version": "0.1.0",
  "seed": 2026,
  "task_count": 12,
  "real_tasks": [
    {
      "task_id": "cifar10_softmax",
      "task_name": "CIFAR-10 softmax",
      "subset": "real",
      "n_samples": 10000,
      "simplex_dim": 10,
      "source_asset": "CIFAR-10",
      "predictor": "Frozen ResNet-18 softmax cache",
      "redistribution": "derived-only",
      "notes": "No raw images are redistributed in this package.",
      "available_arrays": [
        "U",
        "Y",
        "class_names",
        "label_index"
      ]
    },
    {
      "task_id": "topics_20ng",
      "task_name": "20 Newsgroups topics",
      "subset": "real",
      "n_samples": 7539,
      "simplex_dim": 10,
      "source_asset": "20 Newsgroups",
      "predictor": "TF-IDF to topic-mixture kNN regressor",
      "redistribution": "derived-only",
      "notes": "This release exposes only derived simplex arrays, not the raw text corpus.",
      "available_arrays": [
        "U",
        "Y",
        "doc_index"
      ]
    },
    {
      "task_id": "samson_unmixing",
      "task_name": "Samson hyperspectral unmixing",
      "subset": "real",
      "n_samples": 9025,
      "simplex_dim": 3,
      "source_asset": "Samson ROI hyperspectral benchmark",
      "predictor": "NMF abundance estimator",
      "redistribution": "source-cited",
      "notes": "The public bundle did not ship an explicit license file; keep attribution with any downstream reuse.",
      "available_arrays": [
        "U",
        "Y",
        "endmember_names",
        "endmembers",
        "image_shape",
        "pixels"
      ]
    },
    {
      "task_id": "pbmc3k_pseudobulk",
      "task_name": "PBMC3K pseudobulk deconvolution",
      "subset": "real",
      "n_samples": 5000,
      "simplex_dim": 8,
      "source_asset": "PBMC3K single-cell reference",
      "predictor": "NNLS deconvolution from pseudobulk mixtures",
      "redistribution": "derived-only",
      "notes": "This package exports pseudobulk-derived arrays and signatures rather than the raw single-cell matrix.",
      "available_arrays": [
        "U",
        "Y",
        "bulk_expression",
        "cell_type_names",
        "gene_names",
        "signature"
      ]
    },
    {
      "task_id": "utkface_age_ldl",
      "task_name": "UTKFace age label distributions",
      "subset": "real",
      "n_samples": 23687,
      "simplex_dim": 10,
      "source_asset": "UTKFace aligned-and-cropped images",
      "predictor": "Thumbnail feature PCA+kNN age-distribution regressor",
      "redistribution": "derived-only",
      "notes": "The package omits raw face images and keeps only derived features and simplex arrays.",
      "available_arrays": [
        "U",
        "Y",
        "age",
        "image_features"
      ]
    },
    {
      "task_id": "affectivetext_emotions",
      "task_name": "SemEval AffectiveText emotions",
      "subset": "real",
      "n_samples": 1000,
      "simplex_dim": 6,
      "source_asset": "SemEval-2007 Task 14 AffectiveText",
      "predictor": "Frozen zero-shot emotion scorer",
      "redistribution": "derived-only",
      "notes": "The package omits raw headlines and raw API responses, and keeps only derived score arrays.",
      "available_arrays": [
        "U",
        "Y",
        "emotion_names",
        "gold_raw_scores",
        "instance_id",
        "pred_raw_scores"
      ]
    }
  ],
  "synthetic_tasks": [
    {
      "task_id": "d1_homogeneous",
      "task_name": "d1 homogeneous",
      "subset": "synthetic",
      "n_samples": 7500,
      "simplex_dim": 3,
      "source_asset": "Synthetic DGP",
      "predictor": "Oracle mean predictor from the configured DGP",
      "redistribution": "open",
      "regime_label": "Homogeneous negative control",
      "config_file": "configs/synthetic/d1_homogeneous.yaml",
      "seed": 2026,
      "available_arrays": [
        "R",
        "U",
        "X",
        "Y",
        "sigma_true",
        "split"
      ]
    },
    {
      "task_id": "d2_pure_scale",
      "task_name": "d2 pure scale",
      "subset": "synthetic",
      "n_samples": 6250,
      "simplex_dim": 3,
      "source_asset": "Synthetic DGP",
      "predictor": "Oracle mean predictor from the configured DGP",
      "redistribution": "open",
      "regime_label": "Smooth scale heterogeneity",
      "config_file": "configs/synthetic/d2_pure_scale.yaml",
      "seed": 2026,
      "available_arrays": [
        "R",
        "U",
        "X",
        "Y",
        "sigma_true",
        "split"
      ]
    },
    {
      "task_id": "d3_discrete_groups_aligned",
      "task_name": "d3 discrete groups aligned",
      "subset": "synthetic",
      "n_samples": 6250,
      "simplex_dim": 10,
      "source_asset": "Synthetic DGP",
      "predictor": "Oracle mean predictor from the configured DGP",
      "redistribution": "open",
      "regime_label": "Aligned discrete groups",
      "config_file": "configs/synthetic/d3_discrete_groups_aligned.yaml",
      "seed": 2026,
      "available_arrays": [
        "R",
        "U",
        "X",
        "Y",
        "sigma_true",
        "split"
      ]
    },
    {
      "task_id": "d4_model_bias",
      "task_name": "d4 model bias",
      "subset": "synthetic",
      "n_samples": 6250,
      "simplex_dim": 3,
      "source_asset": "Synthetic DGP",
      "predictor": "Oracle mean predictor from the configured DGP",
      "redistribution": "open",
      "regime_label": "Bias-type heterogeneity",
      "config_file": "configs/synthetic/d4_model_bias.yaml",
      "seed": 2026,
      "available_arrays": [
        "R",
        "U",
        "X",
        "Y",
        "sigma_true",
        "split"
      ]
    },
    {
      "task_id": "d5_heavy_tail",
      "task_name": "d5 heavy tail",
      "subset": "synthetic",
      "n_samples": 6250,
      "simplex_dim": 3,
      "source_asset": "Synthetic DGP",
      "predictor": "Oracle mean predictor from the configured DGP",
      "redistribution": "open",
      "regime_label": "Heavy-tail robustness",
      "config_file": "configs/synthetic/d5_heavy_tail.yaml",
      "seed": 2026,
      "available_arrays": [
        "R",
        "U",
        "X",
        "Y",
        "sigma_true",
        "split"
      ]
    },
    {
      "task_id": "d6_high_k",
      "task_name": "d6 high k",
      "subset": "synthetic",
      "n_samples": 17500,
      "simplex_dim": 50,
      "source_asset": "Synthetic DGP",
      "predictor": "Oracle mean predictor from the configured DGP",
      "redistribution": "open",
      "regime_label": "High-dimensional simplex",
      "config_file": "configs/synthetic/d6_high_k.yaml",
      "seed": 2026,
      "available_arrays": [
        "R",
        "U",
        "X",
        "Y",
        "sigma_true",
        "split"
      ]
    }
  ]
}