Function build_default_timeline

training/mlp_train.py:240–303

(dataset_size: int)

def build_default_timeline(dataset_size: int) -> list[TimelineMilestone]:
    if dataset_size <= 0:
        raise ValueError("Dataset must contain at least one example.")

    specs: list[tuple[str, str, str, float, float | None]] = [
        ("initial", "Initial weights", "initial", 0.0, None),
        ("approx_50", "≈50 images", "approx", 50 / BASE_DATASET_SIZE, None),
        ("approx_120", "≈120 images", "approx", 120 / BASE_DATASET_SIZE, None),
        ("approx_250", "≈250 images", "approx", 250 / BASE_DATASET_SIZE, None),
        ("approx_500", "≈500 images", "approx", 500 / BASE_DATASET_SIZE, None),
        ("approx_1k", "≈1k images", "approx", 1_000 / BASE_DATASET_SIZE, None),
        ("approx_2k", "≈2k images", "approx", 2_000 / BASE_DATASET_SIZE, None),
        ("approx_3_5k", "≈3.5k images", "approx", 3_500 / BASE_DATASET_SIZE, None),
        ("approx_5_8k", "≈5.8k images", "approx", 5_800 / BASE_DATASET_SIZE, None),
        ("approx_8_7k", "≈8.7k images", "approx", 8_700 / BASE_DATASET_SIZE, None),
        ("approx_13k", "≈13k images", "approx", 13_000 / BASE_DATASET_SIZE, None),
        ("approx_19_5k", "≈19.5k images", "approx", 19_500 / BASE_DATASET_SIZE, None),
        ("approx_28_5k", "≈28.5k images", "approx", 28_500 / BASE_DATASET_SIZE, None),
        ("approx_40k", "≈40k images", "approx", 40_000 / BASE_DATASET_SIZE, None),
        ("dataset_1x", "1× dataset", "dataset_multiple", 1.0, 1.0),
        ("approx_80k", "≈80k images", "approx", 80_000 / BASE_DATASET_SIZE, None),
        ("dataset_1_5x", "1.5× dataset", "dataset_multiple", 1.5, 1.5),
        ("dataset_2x", "2× dataset", "dataset_multiple", 2.0, 2.0),
        ("dataset_2_5x", "2.5× dataset", "dataset_multiple", 2.5, 2.5),
        ("dataset_3x", "3× dataset", "dataset_multiple", 3.0, 3.0),
        ("dataset_4x", "4× dataset", "dataset_multiple", 4.0, 4.0),
        ("dataset_5x", "5× dataset", "dataset_multiple", 5.0, 5.0),
        ("dataset_6_5x", "6.5× dataset", "dataset_multiple", 6.5, 6.5),
        ("dataset_8_5x", "8.5× dataset", "dataset_multiple", 8.5, 8.5),
        ("dataset_10x", "10× dataset", "dataset_multiple", 10.0, 10.0),
        ("dataset_12_5x", "12.5× dataset", "dataset_multiple", 12.5, 12.5),
        ("dataset_15x", "15× dataset", "dataset_multiple", 15.0, 15.0),
        ("dataset_17_5x", "17.5× dataset", "dataset_multiple", 17.5, 17.5),
        ("dataset_20x", "20× dataset", "dataset_multiple", 20.0, 20.0),
        ("dataset_25x", "25× dataset", "dataset_multiple", 25.0, 25.0),
        ("dataset_30x", "30× dataset", "dataset_multiple", 30.0, 30.0),
        ("dataset_35x", "35× dataset", "dataset_multiple", 35.0, 35.0),
        ("dataset_40x", "40× dataset", "dataset_multiple", 40.0, 40.0),
        ("dataset_45x", "45× dataset", "dataset_multiple", 45.0, 45.0),
        ("dataset_50x", "50× dataset", "dataset_multiple", 50.0, 50.0),
    ]

    milestones: list[TimelineMilestone] = []
    last_threshold = -1
    for identifier, label, kind, ratio, dataset_multiple in specs:
        if ratio <= 0.0:
            threshold_images = 0
        else:
            scaled = dataset_size * ratio
            threshold_images = max(1, int(round(scaled)))
        if threshold_images <= last_threshold:
            threshold_images = last_threshold + 1
        milestones.append(
            TimelineMilestone(
                identifier,
                threshold_images,
                label,
                kind,