{
  "id": 116566,
  "title": "Anyone getting deterministic results with TensorFlow 2.0 or new Keras",
  "url": "/competitions/tensorflow2-question-answering/discussion/116566",
  "author_name": "",
  "post_date": "2019-11-10T01:49:06.563332300Z",
  "votes": 4,
  "comment_count": 5,
  "views": 0,
  "content": "<p>Hi kagglers,\nI just joined this competition and I'm wondering if I should use Torch or TF 2.0.\nIn past kaggle competitions we found out that we can't get determinitistic results with old Keras but we can with Torch.  See <a href=\"https://www.kaggle.com/c/aptos2019-blindness-detection/discussion/100251\">How to get deterministic results in Keras models</a> for more details.</p>\n\n<p>So I'm wondering if anyone is getting deterministic results with TensorFlow 2.0 or new Keras?</p>\n\n<p>thanks!</p>",
  "messages": [
    {
      "id": "669397",
      "postDate": "11/10/2019 01:49:06",
      "content": "<p>Hi kagglers,\nI just joined this competition and I'm wondering if I should use Torch or TF 2.0.\nIn past kaggle competitions we found out that we can't get determinitistic results with old Keras but we can with Torch.  See <a href=\"https://www.kaggle.com/c/aptos2019-blindness-detection/discussion/100251\">How to get deterministic results in Keras models</a> for more details.</p>\n\n<p>So I'm wondering if anyone is getting deterministic results with TensorFlow 2.0 or new Keras?</p>\n\n<p>thanks!</p>",
      "rawMarkdown": "Hi kagglers,\nI just joined this competition and I'm wondering if I should use Torch or TF 2.0.\nIn past kaggle competitions we found out that we can't get determinitistic results with old Keras but we can with Torch.  See [How to get deterministic results in Keras models](https://www.kaggle.com/c/aptos2019-blindness-detection/discussion/100251) for more details.\n\nSo I'm wondering if anyone is getting deterministic results with TensorFlow 2.0 or new Keras?\n\nthanks!",
      "votes": null
    },
    {
      "id": "687327",
      "postDate": "12/04/2019 08:56:39",
      "content": "<p>Update: I switched to Torch and getting deterministic results.</p>",
      "rawMarkdown": "Update: I switched to Torch and getting deterministic results.",
      "votes": null
    },
    {
      "id": "687457",
      "postDate": "12/04/2019 11:35:17",
      "content": "<p>So far I wasn't able to get deterministic results even with all the seeds</p>",
      "rawMarkdown": "So far I wasn't able to get deterministic results even with all the seeds",
      "votes": null
    },
    {
      "id": "687771",
      "postDate": "12/04/2019 19:35:32",
      "content": "<p>tf cpu and torch can have deterministic result, while tf-gpu and tf-tpu can't</p>",
      "rawMarkdown": "tf cpu and torch can have deterministic result, while tf-gpu and tf-tpu can't",
      "votes": null
    },
    {
      "id": "688181",
      "postDate": "12/05/2019 10:14:02",
      "content": "<p>Thank you! That was my guess.</p>",
      "rawMarkdown": "Thank you! That was my guess.",
      "votes": null
    },
    {
      "id": "688454",
      "postDate": "12/05/2019 15:53:30",
      "content": "<p>That actually was the main reason I switch from tf1 to torch. Interesting to hear they still haven't fixed that issue</p>",
      "rawMarkdown": "That actually was the main reason I switch from tf1 to torch. Interesting to hear they still haven't fixed that issue",
      "votes": null
    }
  ],
  "comments": [
    {
      "id": 687327,
      "author_name": "higepon",
      "author_url": "",
      "post_date": "12/04/2019 08:56:39",
      "content": "<p>Update: I switched to Torch and getting deterministic results.</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 687457,
      "author_name": "dimitreoliveira",
      "author_url": "",
      "post_date": "12/04/2019 11:35:17",
      "content": "<p>So far I wasn't able to get deterministic results even with all the seeds</p>",
      "votes": null,
      "replies": []
    },
    {
      "id": 687771,
      "author_name": "httpwwwfszyc",
      "author_url": "",
      "post_date": "12/04/2019 19:35:32",
      "content": "<p>tf cpu and torch can have deterministic result, while tf-gpu and tf-tpu can't</p>",
      "votes": null,
      "replies": [
        {
          "id": 688181,
          "author_name": "higepon",
          "author_url": "",
          "post_date": "12/05/2019 10:14:02",
          "content": "<p>Thank you! That was my guess.</p>",
          "votes": null,
          "replies": []
        }
      ]
    },
    {
      "id": 688454,
      "author_name": "christofhenkel",
      "author_url": "",
      "post_date": "12/05/2019 15:53:30",
      "content": "<p>That actually was the main reason I switch from tf1 to torch. Interesting to hear they still haven't fixed that issue</p>",
      "votes": null,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "669397": "Hi kagglers,\nI just joined this competition and I'm wondering if I should use Torch or TF 2.0.\nIn past kaggle competitions we found out that we can't get determinitistic results with old Keras but we can with Torch.  See [How to get deterministic results in Keras models](https://www.kaggle.com/c/aptos2019-blindness-detection/discussion/100251) for more details.\n\nSo I'm wondering if anyone is getting deterministic results with TensorFlow 2.0 or new Keras?\n\nthanks!",
    "687327": "Update: I switched to Torch and getting deterministic results.",
    "687457": "So far I wasn't able to get deterministic results even with all the seeds",
    "687771": "tf cpu and torch can have deterministic result, while tf-gpu and tf-tpu can't",
    "688181": "Thank you! That was my guess.",
    "688454": "That actually was the main reason I switch from tf1 to torch. Interesting to hear they still haven't fixed that issue"
  },
  "source": "meta"
}