{
  "id": 129977,
  "title": "How can I use TPU without DataGenerator or Dataset?",
  "url": "/competitions/flower-classification-with-tpus/discussion/129977",
  "author_name": "",
  "post_date": "2020-02-11T14:31:46.650563100Z",
  "votes": 2,
  "comment_count": 1,
  "views": 0,
  "content": "<p>Can I fit data in Keras just like using GPU using numpy array or pandas dataframe?</p>",
  "messages": [
    {
      "id": "742767",
      "postDate": "02/11/2020 14:31:46",
      "content": "<p>Can I fit data in Keras just like using GPU using numpy array or pandas dataframe?</p>",
      "rawMarkdown": "Can I fit data in Keras just like using GPU using numpy array or pandas dataframe?",
      "votes": null
    },
    {
      "id": "743060",
      "postDate": "02/11/2020 18:48:42",
      "content": "<p>Numpy arrays work but it might not be the fastest way. For models where the data fits in memory, this might not matter. Not sure about pandas.</p>\n\n<p>Give tf.data.Dataset a try. It is the recommended data API in Tensorflow, and I find it to be really nicely designed.</p>",
      "rawMarkdown": "Numpy arrays work but it might not be the fastest way. For models where the data fits in memory, this might not matter. Not sure about pandas.\n\nGive tf.data.Dataset a try. It is the recommended data API in Tensorflow, and I find it to be really nicely designed.",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 743060,
      "author_name": "mgorner",
      "author_url": "",
      "post_date": "02/11/2020 18:48:42",
      "content": "<p>Numpy arrays work but it might not be the fastest way. For models where the data fits in memory, this might not matter. Not sure about pandas.</p>\n\n<p>Give tf.data.Dataset a try. It is the recommended data API in Tensorflow, and I find it to be really nicely designed.</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "742767": "Can I fit data in Keras just like using GPU using numpy array or pandas dataframe?",
    "743060": "Numpy arrays work but it might not be the fastest way. For models where the data fits in memory, this might not matter. Not sure about pandas.\n\nGive tf.data.Dataset a try. It is the recommended data API in Tensorflow, and I find it to be really nicely designed."
  },
  "source": "meta"
}