This is an automated email from the ASF dual-hosted git repository. fmcquillan pushed a commit to branch master in repository https://gitbox.apache.org/repos/asf/madlib.git
The following commit(s) were added to refs/heads/master by this push: new 1fa020b clarify warm start with model selection in user docs 1fa020b is described below commit 1fa020b08c2a4a8971d1957674794894d6c71783 Author: Frank McQuillan <fmcquil...@pivotal.io> AuthorDate: Tue Jan 7 17:33:42 2020 -0800 clarify warm start with model selection in user docs --- .../modules/deep_learning/madlib_keras_fit_multiple_model.sql_in | 4 +++- 1 file changed, 3 insertions(+), 1 deletion(-) diff --git a/src/ports/postgres/modules/deep_learning/madlib_keras_fit_multiple_model.sql_in b/src/ports/postgres/modules/deep_learning/madlib_keras_fit_multiple_model.sql_in index fbf3497..0468942 100644 --- a/src/ports/postgres/modules/deep_learning/madlib_keras_fit_multiple_model.sql_in +++ b/src/ports/postgres/modules/deep_learning/madlib_keras_fit_multiple_model.sql_in @@ -84,7 +84,7 @@ for the training data. For example, you may only want to train models on segments that reside on hosts that are GPU enabled. You can set up the models and hyperparameters to try with the -<a href="group__grp__keras__model__selection.html">Setup +<a href="group__grp__keras__setup__model__selection.html">Setup Model Selection</a> utility to define the unique combinations of model architectures, compile and fit parameters. @@ -1320,6 +1320,8 @@ set the 'warm_start' parameter to TRUE in the fit function. Transfer learning uses initial model state (weights) stored in the 'model_arch_table' - in this case set the 'warm_start' parameter to FALSE in the fit function. +4. Here are some more details on how warm start works. These details are mostly applicable when implementing autoML algorithms on top of MADlib's model selection. In short, the 'model_selection_table' dictates which models get trained and output to the 'model_output_table' and associated summary and info tables. When 'warm_start' is TRUE, models are built for each 'mst_key' in the 'model_selection_table'. If there are prior runs for an 'mst_key' then the weights from that run will be [...] + @anchor background @par Technical Background