{
  "id": 175729,
  "title": "Not a ground-breaking idea, just one way to keep track of experiments.",
  "url": "/competitions/siim-isic-melanoma-classification/discussion/175729",
  "author_name": "Manjesh Gupta",
  "post_date": "2020-08-19T07:54:49.655000",
  "votes": 14,
  "comment_count": 15,
  "views": 0,
  "content": "<p>Experimentation is integral to data science and one must keep track of all experiments. I am sure everyone has their own way.</p>\n<p>This is how I do it. Sharing here because I think that this might give some insight to beginners (like me) about how to keep track of experiments and <strong>choosing those final submissions</strong>.</p>\n<p>[*These are my results for this competition.]</p>",
  "messages": [
    {
      "id": 976973,
      "postDate": "2020-08-19T07:54:49.657Z",
      "content": "<p>Experimentation is integral to data science and one must keep track of all experiments. I am sure everyone has their own way.</p>\n<p>This is how I do it. Sharing here because I think that this might give some insight to beginners (like me) about how to keep track of experiments and <strong>choosing those final submissions</strong>.</p>\n<p>[*These are my results for this competition.]</p>",
      "rawMarkdown": "Experimentation is integral to data science and one must keep track of all experiments. I am sure everyone has their own way.\n\nThis is how I do it. Sharing here because I think that this might give some insight to beginners (like me) about how to keep track of experiments and **choosing those final submissions**.\n\n[*These are my results for this competition.]",
      "votes": 14
    },
    {
      "id": 977044,
      "postDate": "2020-08-19T08:56:05.037Z",
      "content": "<p>I found a step-change in my ability to run experiments and properly engage with competitions when I started using an experiment tracking software…</p>\n<p>Can't recommend <a href=\"https://neptune.ai/\" target=\"_blank\">neptune.ai</a> enough, although its not great for analyzing the results of your experiments, it is great for logging! (I have no affiliation with neptune) </p>",
      "rawMarkdown": "I found a step-change in my ability to run experiments and properly engage with competitions when I started using an experiment tracking software...\n\nCan't recommend [neptune.ai](https://neptune.ai/) enough, although its not great for analyzing the results of your experiments, it is great for logging! (I have no affiliation with neptune) ",
      "votes": 3,
      "replies": [
        {
          "id": 977183,
          "postDate": "2020-08-19T10:39:37.733Z",
          "content": "<p>Thank you for sharing! Will look into it.</p>",
          "rawMarkdown": "Thank you for sharing! Will look into it."
        },
        {
          "id": 978235,
          "postDate": "2020-08-20T03:40:01.623Z",
          "rawMarkdown": "",
          "isDeleted": true
        }
      ]
    },
    {
      "id": 977022,
      "postDate": "2020-08-19T08:34:23.803Z",
      "content": "<p>Indeed tracking experiments is crucial - btw there are many automated tools that can do that for you, see MLflow, mlrun, determined, Weights &amp; Biases and more. I've used as well in the beginning excel sheets but after some point getting too messy. Now I'm more in to use MLflow and W&amp;B. To get an idea check these articles</p>\n<ul>\n<li><a href=\"https://towardsdatascience.com/tracking-ml-experiments-using-mlflow-7910197091bb\" target=\"_blank\">towardsdatascience - MLflow</a></li>\n<li><a href=\"https://towardsdatascience.com/machine-learning-experiment-tracking-93b796e501b0\" target=\"_blank\">towardsdatascience - W&amp;B</a> </li>\n</ul>\n<p>PS: If you are interested I have also a (public) tutorial notebook that shows step-by-step the process with W&amp;B. </p>\n<p>Edit: Just to be clear, I have no affiliation with either MLflow, W&amp;B - just use this tools with a free account :) </p>",
      "rawMarkdown": "Indeed tracking experiments is crucial - btw there are many automated tools that can do that for you, see MLflow, mlrun, determined, Weights & Biases and more. I've used as well in the beginning excel sheets but after some point getting too messy. Now I'm more in to use MLflow and W&B. To get an idea check these articles\n- [towardsdatascience - MLflow](https://towardsdatascience.com/tracking-ml-experiments-using-mlflow-7910197091bb)\n- [towardsdatascience - W&B](https://towardsdatascience.com/machine-learning-experiment-tracking-93b796e501b0) \n\nPS: If you are interested I have also a (public) tutorial notebook that shows step-by-step the process with W&B. \n\nEdit: Just to be clear, I have no affiliation with either MLflow, W&B - just use this tools with a free account :) ",
      "votes": 3,
      "replies": [
        {
          "id": 977032,
          "postDate": "2020-08-19T08:48:24.153Z",
          "content": "<p>I would add that <a href=\"https://guild.ai/\" target=\"_blank\">guild</a> and <a href=\"https://github.com/IDSIA/sacred\" target=\"_blank\">sacred</a> are also worth looking at</p>",
          "rawMarkdown": "I would add that [guild](https://guild.ai/) and [sacred](https://github.com/IDSIA/sacred) are also worth looking at",
          "votes": 3
        },
        {
          "id": 977042,
          "postDate": "2020-08-19T08:54:10.490Z",
          "content": "<p>Thanks for sharing, I will surely look into these.</p>",
          "rawMarkdown": "Thanks for sharing, I will surely look into these.",
          "votes": 1
        },
        {
          "id": 977105,
          "postDate": "2020-08-19T09:53:49.543Z",
          "content": "<p>I was looking for something like this. Thanks.!</p>",
          "rawMarkdown": "I was looking for something like this. Thanks.!",
          "votes": 1
        }
      ]
    },
    {
      "id": 976979,
      "postDate": "2020-08-19T08:01:00.623Z",
      "content": "<p>Thanks for sharing.  I use spreadsheets too!  I track everything though its quite tedious……….I agree that keeping track is key!  Especially if you want to mine the value in your OOF for stacking etc,</p>",
      "rawMarkdown": "Thanks for sharing.  I use spreadsheets too!  I track everything though its quite tedious..........I agree that keeping track is key!  Especially if you want to mine the value in your OOF for stacking etc,",
      "votes": 3
    },
    {
      "id": 977206,
      "postDate": "2020-08-19T10:53:33.600Z",
      "content": "<p>Thanks for sharing :) I use <code>PAPER + PEN</code> </p>",
      "rawMarkdown": "Thanks for sharing :) I use `PAPER + PEN` ",
      "votes": 4,
      "replies": [
        {
          "id": 977238,
          "postDate": "2020-08-19T11:15:50.640Z",
          "content": "<p>Could you explain more <code>PAPER + PEN</code> or share some articles? :)<br>\n<a href=\"https://www.kaggle.com/coolcoder22\" target=\"_blank\">@coolcoder22</a> </p>",
          "rawMarkdown": "Could you explain more `PAPER + PEN` or share some articles? :)\n@coolcoder22 ",
          "votes": 2
        },
        {
          "id": 977241,
          "postDate": "2020-08-19T11:18:00.240Z",
          "content": "<p>Old school. Whatever works I guess. :)</p>",
          "rawMarkdown": "Old school. Whatever works I guess. :)"
        },
        {
          "id": 977251,
          "postDate": "2020-08-19T11:28:11.423Z",
          "content": "<p><a href=\"https://www.kaggle.com/piantic\" target=\"_blank\">@piantic</a> I write all my experiments on paper </p>",
          "rawMarkdown": "@piantic I write all my experiments on paper ",
          "votes": 1
        }
      ]
    },
    {
      "id": 977126,
      "postDate": "2020-08-19T09:59:27.923Z",
      "content": "<p><a href=\"https://www.kaggle.com/manjeshg03\" target=\"_blank\">@manjeshg03</a> Thanks for sharing and it is a great job.</p>",
      "rawMarkdown": "@manjeshg03 Thanks for sharing and it is a great job.",
      "votes": 1
    },
    {
      "id": 977283,
      "postDate": "2020-08-19T11:46:43.817Z",
      "content": "<p>Thanks for sharing. I use Github <code>Readme.md</code> file to keep track of experiments.<br>\n<img src=\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F2907842%2Fbe2b82573f548c0c7f3438dec21b1985%2FCapture.png?generation=1597837437757549&amp;alt=media\" alt=\"\">.</p>",
      "rawMarkdown": "Thanks for sharing. I use Github `Readme.md` file to keep track of experiments.\n![](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F2907842%2Fbe2b82573f548c0c7f3438dec21b1985%2FCapture.png?generation=1597837437757549&alt=media).",
      "votes": 2
    },
    {
      "id": 977102,
      "postDate": "2020-08-19T09:53:09.480Z",
      "rawMarkdown": "",
      "isDeleted": true
    }
  ],
  "comments": [
    {
      "id": 977044,
      "author_name": "FChmiel",
      "author_url": "",
      "post_date": "2020-08-19T08:56:05.037000",
      "content": "<p>I found a step-change in my ability to run experiments and properly engage with competitions when I started using an experiment tracking software…</p>\n<p>Can't recommend <a href=\"https://neptune.ai/\" target=\"_blank\">neptune.ai</a> enough, although its not great for analyzing the results of your experiments, it is great for logging! (I have no affiliation with neptune) </p>",
      "votes": 3,
      "replies": [
        {
          "id": 977183,
          "author_name": "Manjesh Gupta",
          "author_url": "",
          "post_date": "2020-08-19T10:39:37.733000",
          "content": "<p>Thank you for sharing! Will look into it.</p>",
          "votes": 0,
          "replies": []
        },
        {
          "id": 978235,
          "author_name": "",
          "author_url": "",
          "post_date": "2020-08-20T03:40:01.623000",
          "content": "",
          "votes": 0,
          "replies": []
        }
      ]
    },
    {
      "id": 977022,
      "author_name": "Ioannis M",
      "author_url": "",
      "post_date": "2020-08-19T08:34:23.803000",
      "content": "<p>Indeed tracking experiments is crucial - btw there are many automated tools that can do that for you, see MLflow, mlrun, determined, Weights &amp; Biases and more. I've used as well in the beginning excel sheets but after some point getting too messy. Now I'm more in to use MLflow and W&amp;B. To get an idea check these articles</p>\n<ul>\n<li><a href=\"https://towardsdatascience.com/tracking-ml-experiments-using-mlflow-7910197091bb\" target=\"_blank\">towardsdatascience - MLflow</a></li>\n<li><a href=\"https://towardsdatascience.com/machine-learning-experiment-tracking-93b796e501b0\" target=\"_blank\">towardsdatascience - W&amp;B</a> </li>\n</ul>\n<p>PS: If you are interested I have also a (public) tutorial notebook that shows step-by-step the process with W&amp;B. </p>\n<p>Edit: Just to be clear, I have no affiliation with either MLflow, W&amp;B - just use this tools with a free account :) </p>",
      "votes": 3,
      "replies": [
        {
          "id": 977032,
          "author_name": "Sam Klein",
          "author_url": "",
          "post_date": "2020-08-19T08:48:24.153000",
          "content": "<p>I would add that <a href=\"https://guild.ai/\" target=\"_blank\">guild</a> and <a href=\"https://github.com/IDSIA/sacred\" target=\"_blank\">sacred</a> are also worth looking at</p>",
          "votes": 3,
          "replies": []
        },
        {
          "id": 977042,
          "author_name": "Manjesh Gupta",
          "author_url": "",
          "post_date": "2020-08-19T08:54:10.490000",
          "content": "<p>Thanks for sharing, I will surely look into these.</p>",
          "votes": 1,
          "replies": []
        },
        {
          "id": 977105,
          "author_name": "Prateek Mishra",
          "author_url": "",
          "post_date": "2020-08-19T09:53:49.543000",
          "content": "<p>I was looking for something like this. Thanks.!</p>",
          "votes": 1,
          "replies": []
        }
      ]
    },
    {
      "id": 976979,
      "author_name": "Signal",
      "author_url": "",
      "post_date": "2020-08-19T08:01:00.623000",
      "content": "<p>Thanks for sharing.  I use spreadsheets too!  I track everything though its quite tedious……….I agree that keeping track is key!  Especially if you want to mine the value in your OOF for stacking etc,</p>",
      "votes": 3,
      "replies": []
    },
    {
      "id": 977206,
      "author_name": "Hitesh Gorana",
      "author_url": "",
      "post_date": "2020-08-19T10:53:33.600000",
      "content": "<p>Thanks for sharing :) I use <code>PAPER + PEN</code> </p>",
      "votes": 4,
      "replies": [
        {
          "id": 977238,
          "author_name": "Heroseo",
          "author_url": "",
          "post_date": "2020-08-19T11:15:50.640000",
          "content": "<p>Could you explain more <code>PAPER + PEN</code> or share some articles? :)<br>\n<a href=\"https://www.kaggle.com/coolcoder22\" target=\"_blank\">@coolcoder22</a> </p>",
          "votes": 2,
          "replies": []
        },
        {
          "id": 977241,
          "author_name": "Manjesh Gupta",
          "author_url": "",
          "post_date": "2020-08-19T11:18:00.240000",
          "content": "<p>Old school. Whatever works I guess. :)</p>",
          "votes": 0,
          "replies": []
        },
        {
          "id": 977251,
          "author_name": "Hitesh Gorana",
          "author_url": "",
          "post_date": "2020-08-19T11:28:11.423000",
          "content": "<p><a href=\"https://www.kaggle.com/piantic\" target=\"_blank\">@piantic</a> I write all my experiments on paper </p>",
          "votes": 1,
          "replies": []
        }
      ]
    },
    {
      "id": 977126,
      "author_name": "Heroseo",
      "author_url": "",
      "post_date": "2020-08-19T09:59:27.923000",
      "content": "<p><a href=\"https://www.kaggle.com/manjeshg03\" target=\"_blank\">@manjeshg03</a> Thanks for sharing and it is a great job.</p>",
      "votes": 1,
      "replies": []
    },
    {
      "id": 977283,
      "author_name": "Vatsal Parsaniya",
      "author_url": "",
      "post_date": "2020-08-19T11:46:43.817000",
      "content": "<p>Thanks for sharing. I use Github <code>Readme.md</code> file to keep track of experiments.<br>\n<img src=\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F2907842%2Fbe2b82573f548c0c7f3438dec21b1985%2FCapture.png?generation=1597837437757549&amp;alt=media\" alt=\"\">.</p>",
      "votes": 2,
      "replies": []
    },
    {
      "id": 977102,
      "author_name": "",
      "author_url": "",
      "post_date": "2020-08-19T09:53:09.480000",
      "content": "",
      "votes": 0,
      "replies": []
    }
  ],
  "raw_markdown_by_id": {
    "976973": "Experimentation is integral to data science and one must keep track of all experiments. I am sure everyone has their own way.\n\nThis is how I do it. Sharing here because I think that this might give some insight to beginners (like me) about how to keep track of experiments and **choosing those final submissions**.\n\n[*These are my results for this competition.]",
    "977044": "I found a step-change in my ability to run experiments and properly engage with competitions when I started using an experiment tracking software...\n\nCan't recommend [neptune.ai](https://neptune.ai/) enough, although its not great for analyzing the results of your experiments, it is great for logging! (I have no affiliation with neptune) ",
    "977022": "Indeed tracking experiments is crucial - btw there are many automated tools that can do that for you, see MLflow, mlrun, determined, Weights & Biases and more. I've used as well in the beginning excel sheets but after some point getting too messy. Now I'm more in to use MLflow and W&B. To get an idea check these articles\n- [towardsdatascience - MLflow](https://towardsdatascience.com/tracking-ml-experiments-using-mlflow-7910197091bb)\n- [towardsdatascience - W&B](https://towardsdatascience.com/machine-learning-experiment-tracking-93b796e501b0) \n\nPS: If you are interested I have also a (public) tutorial notebook that shows step-by-step the process with W&B. \n\nEdit: Just to be clear, I have no affiliation with either MLflow, W&B - just use this tools with a free account :) ",
    "976979": "Thanks for sharing.  I use spreadsheets too!  I track everything though its quite tedious..........I agree that keeping track is key!  Especially if you want to mine the value in your OOF for stacking etc,",
    "977206": "Thanks for sharing :) I use `PAPER + PEN` ",
    "977126": "@manjeshg03 Thanks for sharing and it is a great job.",
    "977283": "Thanks for sharing. I use Github `Readme.md` file to keep track of experiments.\n![](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F2907842%2Fbe2b82573f548c0c7f3438dec21b1985%2FCapture.png?generation=1597837437757549&alt=media).",
    "977102": ""
  }
}