hyper parameter optimization

User 597 | 8/11/2014, 3:18:49 PM

for the graphLab create api, is the parameter search techniques (grid search, randomized search) already included and utilized?

Comments

User 18 | 8/12/2014, 5:45:46 PM

graphlab.toolkits.modelparamssearch() uses grid search to tune models. See the API doc (via 'help(func)' or 'func?') for more information. A more detailed user guide is forthcoming.


User 16 | 8/12/2014, 6:18:54 PM

Hi Chan,

Hyper parameter search is included in the most recent version of GraphLab Create. The function is located at <a href="http://graphlab.com/products/create/docs/graphlab.toolkits.modelparametersearch.html">graphlab.toolkits.modelparametersearch</a>.

By default it uses grid search. If you specify a value for maxnummodels, it will uses random search.

Hope that helps. Let me know if you have any other questions.

Toby


User 875 | 4/20/2015, 7:26:22 PM

In the case of recommender: how is it possible to change the list length from 5 to 10 or 12, with other words "trainingprecision@5" to "trainingprecision@10" or "training_precision@12"


User 875 | 4/20/2015, 7:26:22 PM

In the case of recommender: how is it possible to change the list length from 5 to 10 or 12, with other words "trainingprecision@5" to "trainingprecision@10" or "training_precision@12"