René's URL Explorer Experiment

{"@context":"https://schema.org","@type":"DiscussionForumPosting","headline":"vector_of_kll_floats_sketches.get_quantiles() returns wrong values with float32","articleBody":"\n\n```python\n#!/usr/bin/env python3\n\"\"\"\nMinimal example: vector_of_kll_floats_sketches.get_quantiles() returns WRONG VALUES with float32\n\"\"\"\nimport numpy as np\nfrom datasketches import vector_of_kll_floats_sketches\n\n# Create test data: 1000 samples between -100 and -10\nnp.random.seed(42)\ntest_data = np.random.uniform(-100, -10, size=(1000, 1)).astype(np.float32)\n\nprint(\"Test data: 1000 samples between -100 and -10\")\nprint(f\"True min: {test_data.min():.2f}, True max: {test_data.max():.2f}\")\n\n# Create sketch and add data\nkll = vector_of_kll_floats_sketches(200, 1)\nkll.update(test_data)\n\n# Request p0.0001 (should be ~-100) and p0.9999 (should be ~-10)\nranks_list = [0.0001, 0.9999]\nranks_array32 = np.array(ranks_list, dtype=np.float32)\nranks_array64 = np.array(ranks_list, dtype=np.float64)\n\n\nprint(\"\\n\" + \"=\"*60)\nprint(\"BUG: numpy array with dtype=np.float32 returns WRONG quantiles\")\nprint(\"=\"*60)\n\nquants_array = kll.get_quantiles(ranks_array32)\nprint(f\"\\nWith numpy array with dtype=np.float32: {ranks_array32}\")\nprint(f\"  p0.0001 = {quants_array[0][0]:.2f}  (expected: ~-100)\")\nprint(f\"  p0.9999 = {quants_array[0][1]:.2f}  (expected: ~-10)\")\nprint(f\"  ✗ WRONG: Both values near minimum!\")\n\nquants_array64 = kll.get_quantiles(ranks_array64)\nprint(f\"\\nWith numpy array with dtype=np.float64: {ranks_array64}\")\nprint(f\"  p0.0001 = {quants_array64[0][0]:.2f}  (expected: ~-100)\")\nprint(f\"  p0.9999 = {quants_array64[0][1]:.2f}  (expected: ~-10)\")\nprint(f\"  ✓ CORRECT\")\n\n```\n\n```\nTest data: 1000 samples between -100 and -10\nTrue min: -99.58, True max: -10.03\n\n============================================================\nBUG: numpy array with dtype=np.float32 returns WRONG quantiles\n============================================================\n\nWith numpy array with dtype=np.float32: [1.000e-04 9.999e-01]\n  p0.0001 = -98.69  (expected: ~-100)\n  p0.9999 = -99.50  (expected: ~-10)\n  ✗ WRONG: Both values near minimum!\n\nWith numpy array with dtype=np.float64: [1.000e-04 9.999e-01]\n  p0.0001 = -99.50  (expected: ~-100)\n  p0.9999 = -10.28  (expected: ~-10)\n  ✓ CORRECT\n```","author":{"url":"https://github.com/tyler-rt","@type":"Person","name":"tyler-rt"},"datePublished":"2025-12-01T07:45:29.000Z","interactionStatistic":{"@type":"InteractionCounter","interactionType":"https://schema.org/CommentAction","userInteractionCount":2},"url":"https://github.com/63/datasketches-python/issues/63"}