Difference between revisions of "...predict values"
m |
|||
| Line 12: | Line 12: | ||
* ...fit where small deviations are not penalized, then try [[Support Vector Regression (SVR)]] | * ...fit where small deviations are not penalized, then try [[Support Vector Regression (SVR)]] | ||
* ...rank ordered categories, then try the [[Ordinal Regression]] | * ...rank ordered categories, then try the [[Ordinal Regression]] | ||
| − | * ...predict event counts, then try the [[Poisson Regression]] | + | * ...predict event counts, then try the [[Poisson Regression]]; log-linear |
| − | * ...predict a distribution with labeled data, then try the [[Fast Forest Quantile Regression]] | + | * ...predict a distribution rather than point prediction with labeled data, then try the [[Fast Forest Quantile Regression]] |
* ...accuracy matters, I can NOT accept a long training time, and have little memory, then try a [[Decision Forest Regression]] | * ...accuracy matters, I can NOT accept a long training time, and have little memory, then try a [[Decision Forest Regression]] | ||
| − | * ...accuracy matters, I can NOT accept a long training time, then try a [[Boosted Decision Tree Regression]] | + | * ...accuracy matters, I can NOT accept a long training time, then try a [[Boosted Decision Tree Regression]] with large memory fooprint |
* ...accuracy matters, I can accept a long training time, then try a [[General Regression Neural Network (GRNN)]] | * ...accuracy matters, I can accept a long training time, then try a [[General Regression Neural Network (GRNN)]] | ||
Revision as of 20:47, 2 June 2018
When a value is being predicted, as with stock prices, supervised learning is called regression.
___________________________________
I want to...
- ...be fast, nice line fittings, then try the Linear Regression
- ...small dataset, nice line fittings, then try the Bayesian Linear Regression
- ...fit where small deviations are not penalized, then try Support Vector Regression (SVR)
- ...rank ordered categories, then try the Ordinal Regression
- ...predict event counts, then try the Poisson Regression; log-linear
- ...predict a distribution rather than point prediction with labeled data, then try the Fast Forest Quantile Regression
- ...accuracy matters, I can NOT accept a long training time, and have little memory, then try a Decision Forest Regression
- ...accuracy matters, I can NOT accept a long training time, then try a Boosted Decision Tree Regression with large memory fooprint
- ...accuracy matters, I can accept a long training time, then try a General Regression Neural Network (GRNN)
