{ "id": 396643, "title": "What's the best approach to solve this problem?", "url": "/competitions/icecube-neutrinos-in-deep-ice/discussion/396643", "author_name": "", "post_date": "2023-03-22T12:19:49.712175Z", "votes": -4, "comment_count": 2, "views": 0, "content": "

Anyone who can explain in simple terms, how to start solving? How exactly you are connecting the dots?

", "messages": [ { "id": "2192088", "postDate": "03/22/2023 12:19:49", "content": "

Anyone who can explain in simple terms, how to start solving? How exactly you are connecting the dots?

", "rawMarkdown": "Anyone who can explain in simple terms, how to start solving? How exactly you are connecting the dots?", "votes": null }, { "id": "2193015", "postDate": "03/23/2023 03:39:19", "content": "

There are basically two ways:

\n
    \n

  1. Using trodden paths: Look at the solutions others have provided so far and try to find tweaks and improvements wherever possible.

    2. \n

  2. Using new paths: If you think that the solutions so far appear to be very inefficient, try to find something completely different.

    3. \n
\n

In order to find your path, you have to fully comprehend the problem. And that means, you probably want to study the papers that have been attached by the competition host, and more papers related to the problem. After studying the papers, assess your knowledge of algorithms, your creativity and the time frame to solve. Walking a new path might be more time-intensive.

\n

Happy solving!

", "rawMarkdown": "There are basically two ways:\n\n1. Using trodden paths: Look at the solutions others have provided so far and try to find tweaks and improvements wherever possible.\n\n2. Using new paths: If you think that the solutions so far appear to be very inefficient, try to find something completely different.\n\nIn order to find your path, you have to fully comprehend the problem. And that means, you probably want to study the papers that have been attached by the competition host, and more papers related to the problem. After studying the papers, assess your knowledge of algorithms, your creativity and the time frame to solve. Walking a new path might be more time-intensive.\n\nHappy solving!", "votes": null }, { "id": "2193084", "postDate": "03/23/2023 04:48:08", "content": "

Using Stephan's #1 path might be a good idea if you have access to lots of compute. For the models I have played with, running more of the data always has improved the result.

\n

I am trying to run all the batches on a dual GPU machine with my version of an LSTM - 5 epochs looks like it's going to take at least a week of run time on my machine and I am pretty sure after day 2 of this that 5 epochs is not going to be sufficient.

", "rawMarkdown": "Using [Stephan's ](https://www.kaggle.com/taqseorangpun)#1 path might be a good idea if you have access to lots of compute. For the models I have played with, running more of the data always has improved the result. \n\nI am trying to run all the batches on a dual GPU machine with my version of an LSTM - 5 epochs looks like it's going to take at least a week of run time on my machine and I am pretty sure after day 2 of this that 5 epochs is not going to be sufficient.", "votes": null } ], "comments": [ { "id": 2193015, "author_name": "taqseorangpun", "author_url": "", "post_date": "03/23/2023 03:39:19", "content": "

There are basically two ways:

\n
    \n

  1. Using trodden paths: Look at the solutions others have provided so far and try to find tweaks and improvements wherever possible.

    2. \n

  2. Using new paths: If you think that the solutions so far appear to be very inefficient, try to find something completely different.

    3. \n
\n

In order to find your path, you have to fully comprehend the problem. And that means, you probably want to study the papers that have been attached by the competition host, and more papers related to the problem. After studying the papers, assess your knowledge of algorithms, your creativity and the time frame to solve. Walking a new path might be more time-intensive.

\n

Happy solving!

", "votes": null, "replies": [ { "id": 2193084, "author_name": "pcjimmmy", "author_url": "", "post_date": "03/23/2023 04:48:08", "content": "

Using Stephan's #1 path might be a good idea if you have access to lots of compute. For the models I have played with, running more of the data always has improved the result.

\n

I am trying to run all the batches on a dual GPU machine with my version of an LSTM - 5 epochs looks like it's going to take at least a week of run time on my machine and I am pretty sure after day 2 of this that 5 epochs is not going to be sufficient.

", "votes": null, "replies": [] } ] } ], "raw_markdown_by_id": { "2192088": "Anyone who can explain in simple terms, how to start solving? How exactly you are connecting the dots?", "2193015": "There are basically two ways:\n\n1. Using trodden paths: Look at the solutions others have provided so far and try to find tweaks and improvements wherever possible.\n\n2. Using new paths: If you think that the solutions so far appear to be very inefficient, try to find something completely different.\n\nIn order to find your path, you have to fully comprehend the problem. And that means, you probably want to study the papers that have been attached by the competition host, and more papers related to the problem. After studying the papers, assess your knowledge of algorithms, your creativity and the time frame to solve. Walking a new path might be more time-intensive.\n\nHappy solving!", "2193084": "Using [Stephan's ](https://www.kaggle.com/taqseorangpun)#1 path might be a good idea if you have access to lots of compute. For the models I have played with, running more of the data always has improved the result. \n\nI am trying to run all the batches on a dual GPU machine with my version of an LSTM - 5 epochs looks like it's going to take at least a week of run time on my machine and I am pretty sure after day 2 of this that 5 epochs is not going to be sufficient." }, "source": "meta" }