{
  "id": 320436,
  "title": "Fascinating Competition",
  "url": "/competitions/herbarium-2022-fgvc9/discussion/320436",
  "author_name": "ArnoldRosielle",
  "post_date": "2022-04-21T16:22:11.031000",
  "votes": 2,
  "comment_count": 0,
  "views": 0,
  "content": "<p>Hi - I am still pretty new to ML and this competition has taught me a lot.  Unfortunately I am not familiar yet with Res Net but I am OK with VGG and VGG16.  For those who my be interested here is my approach which I doubt I will ever complete.<br>\nI tried predicting family first - there are 272 of them. Trying to predict all at once seemed hopeless so I decided to create a model for predicting 8 families at a time.  That gave me 34 models.  I then created a dataframe for the 200K+ test rows with the predictions for each model across the columns of the dataframe.  And chose the best.<br>\nI know which genera are compatible with each family so now I am in the process of going through those 200k test rows where I have already predicted family  and training on vaiid genera and predicting the best.  From what I can see on the prediction probabilities so far made it is working out quite well but this will take forever.  And of course if this ever gets completed I'd have to do the same thing for species.  Perhaps a super computer could do this quite quickly - but all I have is a MAC M1.  Never mind - it has still been fun!</p>",
  "messages": [
    {
      "id": 1763575,
      "postDate": "2022-04-21T16:22:11.033Z",
      "content": "<p>Hi - I am still pretty new to ML and this competition has taught me a lot.  Unfortunately I am not familiar yet with Res Net but I am OK with VGG and VGG16.  For those who my be interested here is my approach which I doubt I will ever complete.<br>\nI tried predicting family first - there are 272 of them. Trying to predict all at once seemed hopeless so I decided to create a model for predicting 8 families at a time.  That gave me 34 models.  I then created a dataframe for the 200K+ test rows with the predictions for each model across the columns of the dataframe.  And chose the best.<br>\nI know which genera are compatible with each family so now I am in the process of going through those 200k test rows where I have already predicted family  and training on vaiid genera and predicting the best.  From what I can see on the prediction probabilities so far made it is working out quite well but this will take forever.  And of course if this ever gets completed I'd have to do the same thing for species.  Perhaps a super computer could do this quite quickly - but all I have is a MAC M1.  Never mind - it has still been fun!</p>",
      "rawMarkdown": "Hi - I am still pretty new to ML and this competition has taught me a lot.  Unfortunately I am not familiar yet with Res Net but I am OK with VGG and VGG16.  For those who my be interested here is my approach which I doubt I will ever complete.\nI tried predicting family first - there are 272 of them. Trying to predict all at once seemed hopeless so I decided to create a model for predicting 8 families at a time.  That gave me 34 models.  I then created a dataframe for the 200K+ test rows with the predictions for each model across the columns of the dataframe.  And chose the best.\nI know which genera are compatible with each family so now I am in the process of going through those 200k test rows where I have already predicted family  and training on vaiid genera and predicting the best.  From what I can see on the prediction probabilities so far made it is working out quite well but this will take forever.  And of course if this ever gets completed I'd have to do the same thing for species.  Perhaps a super computer could do this quite quickly - but all I have is a MAC M1.  Never mind - it has still been fun!",
      "votes": 2
    }
  ],
  "comments": [],
  "raw_markdown_by_id": {
    "1763575": "Hi - I am still pretty new to ML and this competition has taught me a lot.  Unfortunately I am not familiar yet with Res Net but I am OK with VGG and VGG16.  For those who my be interested here is my approach which I doubt I will ever complete.\nI tried predicting family first - there are 272 of them. Trying to predict all at once seemed hopeless so I decided to create a model for predicting 8 families at a time.  That gave me 34 models.  I then created a dataframe for the 200K+ test rows with the predictions for each model across the columns of the dataframe.  And chose the best.\nI know which genera are compatible with each family so now I am in the process of going through those 200k test rows where I have already predicted family  and training on vaiid genera and predicting the best.  From what I can see on the prediction probabilities so far made it is working out quite well but this will take forever.  And of course if this ever gets completed I'd have to do the same thing for species.  Perhaps a super computer could do this quite quickly - but all I have is a MAC M1.  Never mind - it has still been fun!"
  }
}