3 ms·
> (especially) techniques for tabular data, start with the machine learning course. Do you think DL techniques are going to better than traditional techniques
by zawerf 8y ago
> (especially) techniques for tabular data, start with the machine learning course.
Do you think DL techniques are going to better than traditional techniques for tabular data in the future?
This is in reference to your article: http://www.fast.ai/2018/04/29/categorical-embeddings/ http://www.fast.ai/2018/04/29/categorical-embeddings/ and lessons 3 and 4 from the DL course.
- jph00 8y agoYes probably. For instance, there's been interesting recent examples from Pinterest and Instacart (amongst others) who found that DL-based approaches for tabular data reduced the engineering and maintenance headaches due to less need for manual feature engineering. Although decision tree approaches maybe will continue to be faster to train in many cases. At the moment, it really helps to be familiar with both approaches.