{
  "id": 599600,
  "title": "The order of columns is important",
  "url": "/competitions/ariel-data-challenge-2025/discussion/599600",
  "author_name": "DennisSakva",
  "post_date": "2025-08-17T16:05:06.735000",
  "votes": 11,
  "comment_count": 0,
  "views": 0,
  "content": "<p>Just learned it the hard way. When using competition metrics code, don’t forget that the order of columns matters. The code relies on integer indexing instead of column names:</p>\n<pre><code>y_pred = submission.iloc[:, :n_wavelengths].values\nsigma_pred = np.clip(submission.iloc[:, n_wavelengths:].values, a_min=**-, a_max=)\n</code></pre>\n<p>So if you decide to interleave wavelengths and sigmas or something like that, you’re going to have a bad time.</p>",
  "messages": [
    {
      "id": 3270925,
      "postDate": "2025-08-17T16:05:06.737Z",
      "content": "<p>Just learned it the hard way. When using competition metrics code, don’t forget that the order of columns matters. The code relies on integer indexing instead of column names:</p>\n<pre><code>y_pred = submission.iloc[:, :n_wavelengths].values\nsigma_pred = np.clip(submission.iloc[:, n_wavelengths:].values, a_min=**-, a_max=)\n</code></pre>\n<p>So if you decide to interleave wavelengths and sigmas or something like that, you’re going to have a bad time.</p>",
      "rawMarkdown": "Just learned it the hard way. When using competition metrics code, don’t forget that the order of columns matters. The code relies on integer indexing instead of column names:\n```python\ny_pred = submission.iloc[:, :n_wavelengths].values\nsigma_pred = np.clip(submission.iloc[:, n_wavelengths:].values, a_min=10**-15, a_max=None)\n\n```So if you decide to interleave wavelengths and sigmas or something like that, you’re going to have a bad time.\n\n",
      "votes": 11
    }
  ],
  "comments": [],
  "raw_markdown_by_id": {
    "3270925": "Just learned it the hard way. When using competition metrics code, don’t forget that the order of columns matters. The code relies on integer indexing instead of column names:\n```python\ny_pred = submission.iloc[:, :n_wavelengths].values\nsigma_pred = np.clip(submission.iloc[:, n_wavelengths:].values, a_min=10**-15, a_max=None)\n\n```So if you decide to interleave wavelengths and sigmas or something like that, you’re going to have a bad time.\n\n"
  }
}