|
706 | 706 | "\n" |
707 | 707 | ] |
708 | 708 | }, |
| 709 | + { |
| 710 | + "cell_type": "markdown", |
| 711 | + "metadata": {}, |
| 712 | + "source": [ |
| 713 | + "### Another similar idea with multiple modes: mixture density network\n", |
| 714 | + "\n", |
| 715 | + "Great sample demonstrating mixture density network for probabilistic regression [here](https://deep-and-shallow.com/2021/03/20/mixture-density-networks-probabilistic-regression-for-uncertainty-estimation).\n", |
| 716 | + "The sample uses [pytorch tabular](https://pytorch-tabular.readthedocs.io/en/latest/) for its implementation, and interestingly, tackles the small tricks used to reduce the chance of mode collapse such as l1/l2 regularization of [Gumbel softmax](https://towardsdatascience.com/what-is-gumbel-softmax-7f6d9cdcb90e)\n", |
| 717 | + "See code [here](https://github.com/manujosephv/pytorch_tabular/tree/main/src/pytorch_tabular/models/mixture_density)\n", |
| 718 | + "and documentatton [here](https://pytorch-tabular.readthedocs.io/en/latest/tutorials/07-Probabilistic%20Regression%20with%20MDN)" |
| 719 | + ] |
| 720 | + }, |
709 | 721 | { |
710 | 722 | "cell_type": "markdown", |
711 | 723 | "metadata": {}, |
|
0 commit comments