[madlib] 08/08: additional user docs updates about installing Dill and Hyperopt

[kh...@apache.org]

This is an automated email from the ASF dual-hosted git repository.

khannaekta pushed a commit to branch master
in repository https://gitbox.apache.org/repos/asf/madlib.git

commit 33c1a6d3a534183952f575dc08365cd822c4837b
Author: Frank McQuillan <fm...@pivotal.io>
AuthorDate: Mon Oct 26 12:56:26 2020 -0700

    additional user docs updates about installing Dill and Hyperopt
---
 .../modules/deep_learning/madlib_keras_automl.sql_in     | 16 ++++++++++++++--
 .../deep_learning/madlib_keras_custom_function.sql_in    |  6 +++++-
 2 files changed, 19 insertions(+), 3 deletions(-)

diff --git a/src/ports/postgres/modules/deep_learning/madlib_keras_automl.sql_in b/src/ports/postgres/modules/deep_learning/madlib_keras_automl.sql_in
index f703cf0..f692f13 100644
--- a/src/ports/postgres/modules/deep_learning/madlib_keras_automl.sql_in
+++ b/src/ports/postgres/modules/deep_learning/madlib_keras_automl.sql_in
@@ -183,7 +183,12 @@ madlib_keras_automl(
 
   <dt>automl_method (optional)</dt>
   <dd>VARCHAR, default 'hyperband'. Name of the autoML algorithm to run.
-  Can be either 'hyperband' or hyperopt' (case insensitive).
+  Can be either 'hyperband' or 'hyperopt' (case insensitive).
+
+  @note
+  If you select 'hyperopt', then the Hyperopt package must be installed on the main node
+  of the database cluster [3]. Hyperband does not need any separate package installation.
+
   </dd>
 
   <dt>automl_params (optional)</dt>
@@ -1313,7 +1318,12 @@ SELECT * FROM iris_predict ORDER BY id;
 @anchor notes
 @par Notes
 
-In practice you may need to do more than one run of an autoML method to arrive
+1. Hyperopt must be installed on the main node of the database cluster
+if you want to use the Hyperopt method of autoML.
+You can pip install it in the usual way [3].  Hyperband does not require
+any separate package installation.
+
+2. In practice you may need to do more than one run of an autoML method to arrive
 at a model with adequate accuracy.  One approach is to set the search space to
 be quite broad initially, then observe which hyperparameter ranges and model architectures
 seem to be doing the best.  Subesquent runs can then zoom in on those good ones
@@ -1330,6 +1340,8 @@ Hyperparameter Optimization in Hundreds of Dimensions for Vision Architectures,"
 <em>Proceedings of the 30th International Conference on Machine Learning</em>, Atlanta, Georgia,
 USA, 2013. JMLR: W&CP volume 28.
 
+[3] Python catalog for Hyperopt https://pypi.org/project/hyperopt/
+
 @anchor related
 @par Related Topics
 
diff --git a/src/ports/postgres/modules/deep_learning/madlib_keras_custom_function.sql_in b/src/ports/postgres/modules/deep_learning/madlib_keras_custom_function.sql_in
index 43f6afc..440d814 100644
--- a/src/ports/postgres/modules/deep_learning/madlib_keras_custom_function.sql_in
+++ b/src/ports/postgres/modules/deep_learning/madlib_keras_custom_function.sql_in
@@ -83,6 +83,10 @@ load_custom_function(
   <dd>BYTEA. PostgreSQL binary data type of the Python object.
   Object must be created with the Dill package for serializing
   Python objects.
+
+  @note
+  The Dill package must be installed on all segments of the
+  database cluster [1].
   </dd>
 
   <dt>name</dt>
@@ -318,7 +322,7 @@ SELECT id, name, description FROM custom_function_table ORDER BY id;
 @anchor literature
 @literature
 
-[1] Dill https://pypi.org/project/dill/
+[1] Python catalog for Dill package https://pypi.org/project/dill/
 
 [2] https://keras.io/api/metrics/accuracy_metrics/#topkcategoricalaccuracy-class