{
  "id": 166502,
  "title": "my vgg-16 kernel",
  "url": "/competitions/prostate-cancer-grade-assessment/discussion/166502",
  "author_name": "andy jennings",
  "post_date": "2020-07-13T06:32:53.769000",
  "votes": 3,
  "comment_count": 0,
  "views": 0,
  "content": "<p>Thanks for all the encouragement for my kernel. It is a bit of a sorry story as I am on a steep learning curve for quite a few things here. There will be a bit of a pause here. After many weeks struggling with the learning phase for Keras, with strange memory problems I finally managed to get it stable by forcing gc.collect() periodically. I split off the testing into a seperate kernel. Now I have very serious memory leaks from repeated execution of keras.predict() I struggle to believe this at first, but then I found a post from a competition winner complaining about this memory problem! Do you people have machines with 200G of memory? So now I have to re-work this in PyTorch. Honestly I am very surprised by all this. Does nobody implement a production system in Keras? </p>",
  "messages": [
    {
      "id": 927007,
      "postDate": "2020-07-13T06:32:53.770Z",
      "content": "<p>Thanks for all the encouragement for my kernel. It is a bit of a sorry story as I am on a steep learning curve for quite a few things here. There will be a bit of a pause here. After many weeks struggling with the learning phase for Keras, with strange memory problems I finally managed to get it stable by forcing gc.collect() periodically. I split off the testing into a seperate kernel. Now I have very serious memory leaks from repeated execution of keras.predict() I struggle to believe this at first, but then I found a post from a competition winner complaining about this memory problem! Do you people have machines with 200G of memory? So now I have to re-work this in PyTorch. Honestly I am very surprised by all this. Does nobody implement a production system in Keras? </p>",
      "rawMarkdown": "Thanks for all the encouragement for my kernel. It is a bit of a sorry story as I am on a steep learning curve for quite a few things here. There will be a bit of a pause here. After many weeks struggling with the learning phase for Keras, with strange memory problems I finally managed to get it stable by forcing gc.collect() periodically. I split off the testing into a seperate kernel. Now I have very serious memory leaks from repeated execution of keras.predict() I struggle to believe this at first, but then I found a post from a competition winner complaining about this memory problem! Do you people have machines with 200G of memory? So now I have to re-work this in PyTorch. Honestly I am very surprised by all this. Does nobody implement a production system in Keras? ",
      "votes": 3
    }
  ],
  "comments": [],
  "raw_markdown_by_id": {
    "927007": "Thanks for all the encouragement for my kernel. It is a bit of a sorry story as I am on a steep learning curve for quite a few things here. There will be a bit of a pause here. After many weeks struggling with the learning phase for Keras, with strange memory problems I finally managed to get it stable by forcing gc.collect() periodically. I split off the testing into a seperate kernel. Now I have very serious memory leaks from repeated execution of keras.predict() I struggle to believe this at first, but then I found a post from a competition winner complaining about this memory problem! Do you people have machines with 200G of memory? So now I have to re-work this in PyTorch. Honestly I am very surprised by all this. Does nobody implement a production system in Keras? "
  }
}