{ "id": 217638, "title": "Has anyone tired pseudo labelling ?", "url": "/competitions/rfcx-species-audio-detection/discussion/217638", "author_name": "", "post_date": "2021-02-07T16:44:57.032414100Z", "votes": 2, "comment_count": 2, "views": 0, "content": "

What is Pseudo Labeling?
\nRetrain your model with predicted test data.
\nprocess.

\n
    \n
  1. Train on the labelled data.
    2. \n
  2. Predict labels for test dataset.
    3. \n
  3. create pseudo labels on predicted labels
    4. \n
  4. Build a new model and train using pseudo labels
    5. \n
  5. Predict using new model on test dataset and submit to Kaggle
    6. \n
\n

Some Examples
\nNote Book Example
\nPseudo Labeling - QDA - [0.969]
\nYoloV5 Pseudo Labeling

", "messages": [ { "id": "1190335", "postDate": "02/07/2021 16:44:57", "content": "

What is Pseudo Labeling?
\nRetrain your model with predicted test data.
\nprocess.

\n
    \n
  1. Train on the labelled data.
    2. \n
  2. Predict labels for test dataset.
    3. \n
  3. create pseudo labels on predicted labels
    4. \n
  4. Build a new model and train using pseudo labels
    5. \n
  5. Predict using new model on test dataset and submit to Kaggle
    6. \n
\n

Some Examples
\nNote Book Example
\nPseudo Labeling - QDA - [0.969]
\nYoloV5 Pseudo Labeling

", "rawMarkdown": "**What is Pseudo Labeling?**\nRetrain your model with predicted test data. \nprocess. \n1. Train on the labelled data.\n2. Predict labels for test dataset. \n3. create pseudo labels on predicted labels\n4. Build a new model and train using pseudo labels\n5. Predict using new model on test dataset and submit to Kaggle\n\n\n**Some Examples**\n[Note Book Example](https://github.com/stanleyjzheng/PyData/blob/master/exploring-pseudolabelling-schemes-pydata.ipynb) \n[Pseudo Labeling - QDA - [0.969]](https://www.kaggle.com/cdeotte/pseudo-labeling-qda-0-969)\n[YoloV5 Pseudo Labeling](https://www.kaggle.com/nvnnghia/yolov5-pseudo-labeling)", "votes": null }, { "id": "1191668", "postDate": "02/08/2021 15:44:09", "content": "

Not sure if it's a good approach to train on TEST data as the test data is publically available and the same data(a portion of it) will be used for private scoring.
\nA better approach is to psudolabel Training data to find the missing labels and use them for training . but, I had no luck with the missing label. no improvement.

", "rawMarkdown": "Not sure if it's a good approach to train on TEST data as the test data is publically available and the same data(a portion of it) will be used for private scoring. \nA better approach is to psudolabel Training data to find the missing labels and use them for training . but, I had no luck with the missing label. no improvement.", "votes": null }, { "id": "1219034", "postDate": "02/26/2021 11:44:13", "content": "

Look at the Top results. They have used Pseudo Label.

\n

Peace.

", "rawMarkdown": "Look at the Top results. They have used Pseudo Label.\n\n\nPeace.", "votes": null } ], "comments": [ { "id": 1191668, "author_name": "yuvaramsingh", "author_url": "", "post_date": "02/08/2021 15:44:09", "content": "

Not sure if it's a good approach to train on TEST data as the test data is publically available and the same data(a portion of it) will be used for private scoring.
\nA better approach is to psudolabel Training data to find the missing labels and use them for training . but, I had no luck with the missing label. no improvement.

", "votes": null, "replies": [ { "id": 1219034, "author_name": "mlneo07", "author_url": "", "post_date": "02/26/2021 11:44:13", "content": "

Look at the Top results. They have used Pseudo Label.

\n

Peace.

", "votes": null, "replies": [] } ] } ], "raw_markdown_by_id": { "1190335": "**What is Pseudo Labeling?**\nRetrain your model with predicted test data. \nprocess. \n1. Train on the labelled data.\n2. Predict labels for test dataset. \n3. create pseudo labels on predicted labels\n4. Build a new model and train using pseudo labels\n5. Predict using new model on test dataset and submit to Kaggle\n\n\n**Some Examples**\n[Note Book Example](https://github.com/stanleyjzheng/PyData/blob/master/exploring-pseudolabelling-schemes-pydata.ipynb) \n[Pseudo Labeling - QDA - [0.969]](https://www.kaggle.com/cdeotte/pseudo-labeling-qda-0-969)\n[YoloV5 Pseudo Labeling](https://www.kaggle.com/nvnnghia/yolov5-pseudo-labeling)", "1191668": "Not sure if it's a good approach to train on TEST data as the test data is publically available and the same data(a portion of it) will be used for private scoring. \nA better approach is to psudolabel Training data to find the missing labels and use them for training . but, I had no luck with the missing label. no improvement.", "1219034": "Look at the Top results. They have used Pseudo Label.\n\n\nPeace." }, "source": "meta" }