DTree: Update defaults for max_depth, num_splits Reduce the defaults for max_depth to 7 and num_splits to 20 to decrease the chances of running out of memory when initializing tree for problems with many features or with features with many categorical values.
Closes #117 Project: http://git-wip-us.apache.org/repos/asf/incubator-madlib/repo Commit: http://git-wip-us.apache.org/repos/asf/incubator-madlib/commit/3eec0a82 Tree: http://git-wip-us.apache.org/repos/asf/incubator-madlib/tree/3eec0a82 Diff: http://git-wip-us.apache.org/repos/asf/incubator-madlib/diff/3eec0a82 Branch: refs/heads/latest_release Commit: 3eec0a82ee522101264c6557457602f9e0dbee52 Parents: 8faf622 Author: Rahul Iyer <ri...@apache.org> Authored: Tue Apr 18 11:53:36 2017 -0700 Committer: Rahul Iyer <ri...@apache.org> Committed: Tue Apr 18 17:19:54 2017 -0700 ---------------------------------------------------------------------- .../recursive_partitioning/decision_tree.py_in | 23 ++++---- .../recursive_partitioning/decision_tree.sql_in | 59 +++++++++++--------- .../test/decision_tree.sql_in | 2 +- 3 files changed, 46 insertions(+), 38 deletions(-) ---------------------------------------------------------------------- http://git-wip-us.apache.org/repos/asf/incubator-madlib/blob/3eec0a82/src/ports/postgres/modules/recursive_partitioning/decision_tree.py_in ---------------------------------------------------------------------- diff --git a/src/ports/postgres/modules/recursive_partitioning/decision_tree.py_in b/src/ports/postgres/modules/recursive_partitioning/decision_tree.py_in index fb18278..f7c4bd8 100644 --- a/src/ports/postgres/modules/recursive_partitioning/decision_tree.py_in +++ b/src/ports/postgres/modules/recursive_partitioning/decision_tree.py_in @@ -223,7 +223,7 @@ SELECT {schema_madlib}.tree_train( is NULL 'weights', -- A Column name containing weights for each observation. Default is NULL - max_depth, -- Maximum depth of any node, default is 10 + max_depth, -- Maximum depth of any node, default is 7 min_split, -- Minimum number of observations that must exist in a node for a split to be attemped, default is 20 @@ -231,7 +231,7 @@ SELECT {schema_madlib}.tree_train( terminal node, default is min_split/3 n_bins, -- Number of bins to find possible node split threshold values for continuous - variables, default is 100 (Must be greater than 1) + variables, default is 20 (Must be greater than 1) pruning_params, -- A comma-separated text containing key=value pairs of parameters for pruning. Parameters accepted: @@ -341,7 +341,6 @@ def _extract_pruning_params(pruning_params_str): @param pruning_param: str, Parameters used for pruning the tree cp = Cost-complexity for pruning n_folds = Number of folds for cross-validation - Returns: dict. A dictionary containing the pruning parameters """ @@ -567,17 +566,21 @@ def tree_train(schema_madlib, training_table_name, output_table_name, """ msg_level = "notice" if verbose_mode else "warning" - # Set default values for optional arguments - min_split = 20 if (min_split is None and min_bucket is None) else min_split - min_bucket = min_split // 3 if min_bucket is None else min_bucket - min_split = min_bucket * 3 if min_split is None else min_split - n_bins = 100 if n_bins is None else n_bins + # Set default values for all arguments split_criterion = 'gini' if not split_criterion else split_criterion - plpy.notice("split_criterion:" + split_criterion) + max_depth = 7 if max_depth is None else max_depth + if min_split is None and min_bucket is None: + min_split = 20 + min_bucket = 6 + else: + min_bucket = min_split // 3 if min_bucket is None else min_bucket + min_split = min_bucket * 3 if min_split is None else min_split + n_bins = 20 if n_bins is None else n_bins + + # defaults for cp and n_folds set within _extract_pruning_params pruning_param_dict = _extract_pruning_params(pruning_params) cp = pruning_param_dict['cp'] n_folds = pruning_param_dict['n_folds'] - surrogate_param_dict = extract_keyvalue_params(surrogate_params, dict(max_surrogates=int), dict(max_surrogates=0)) http://git-wip-us.apache.org/repos/asf/incubator-madlib/blob/3eec0a82/src/ports/postgres/modules/recursive_partitioning/decision_tree.sql_in ---------------------------------------------------------------------- diff --git a/src/ports/postgres/modules/recursive_partitioning/decision_tree.sql_in b/src/ports/postgres/modules/recursive_partitioning/decision_tree.sql_in index 97e8471..ef671fc 100644 --- a/src/ports/postgres/modules/recursive_partitioning/decision_tree.sql_in +++ b/src/ports/postgres/modules/recursive_partitioning/decision_tree.sql_in @@ -258,7 +258,7 @@ tree_train( <DD>TEXT. Column name containing weights for each observation.</DD> <DT>max_depth (optional)</DT> - <DD>INTEGER, default: 10. Maximum depth of any node of the final tree, + <DD>INTEGER, default: 7. Maximum depth of any node of the final tree, with the root node counted as depth 0.</DD> <DT>min_split (optional)</DT> @@ -272,7 +272,7 @@ tree_train( set to min_bucket*3 or min_bucket to min_split/3, as appropriate.</DD> <DT>num_splits (optional)</DT> - <DD>INTEGER, default: 100. Continuous-valued features are binned into + <DD>INTEGER, default: 20. Continuous-valued features are binned into discrete quantiles to compute split boundaries. This global parameter is used to compute the resolution of splits for continuous features. Higher number of bins will lead to better prediction, @@ -920,7 +920,7 @@ File decision_tree.sql_in documenting the training function * multiple decision trees, one for each group. * @param weights OPTIONAL. Column name containing weights for * each observation. - * @param max_depth OPTIONAL (Default = 10). Set the maximum depth + * @param max_depth OPTIONAL (Default = 7). Set the maximum depth * of any node of the final tree, with the root node counted * as depth 0. * @param min_split OPTIONAL (Default = 20). Minimum number of @@ -931,13 +931,13 @@ File decision_tree.sql_in documenting the training function * one of minbucket or minsplit is specified, minsplit * is set to minbucket*3 or minbucket to minsplit/3, as * appropriate. - * @param n_bins optional (default = 100) number of bins to use + * @param n_bins optional (default = 20) number of bins to use * during binning. continuous-valued features are binned * into discrete bins (per the quartile values) to compute * split bound- aries. this global parameter is used to * compute the resolution of the bins. higher number of * bins will lead to higher processing time. - * @param pruning_params (default = 'cp=0.01') pruning parameter string + * @param pruning_params (default: cp=0) pruning parameter string * containing key-value pairs. * the keys can be: * cp (default = 0.01) a complexity parameter @@ -1574,8 +1574,10 @@ CREATE OR REPLACE FUNCTION MADLIB_SCHEMA.tree_train( pruning_params TEXT, surrogate_params TEXT ) RETURNS VOID AS $$ + -- verbose = false SELECT MADLIB_SCHEMA.tree_train($1, $2, $3, $4, $5, $6, $7, $8, $9, $10, $11, $12, $13, $14, $15, FALSE); + $$ LANGUAGE SQL VOLATILE m4_ifdef(`__HAS_FUNCTION_PROPERTIES__', `MODIFIES SQL DATA', `'); @@ -1596,7 +1598,7 @@ CREATE OR REPLACE FUNCTION MADLIB_SCHEMA.tree_train( pruning_params TEXT ) RETURNS VOID AS $$ SELECT MADLIB_SCHEMA.tree_train($1, $2, $3, $4, $5, $6, $7, $8, $9, $10, - $11, $12, $13, $14, 'max_surrogates=0', FALSE); + $11, $12, $13, $14, NULL::text, FALSE); $$ LANGUAGE SQL VOLATILE m4_ifdef(`__HAS_FUNCTION_PROPERTIES__', `MODIFIES SQL DATA', `'); @@ -1616,8 +1618,8 @@ CREATE OR REPLACE FUNCTION MADLIB_SCHEMA.tree_train( n_bins INTEGER ) RETURNS VOID AS $$ SELECT MADLIB_SCHEMA.tree_train($1, $2, $3, $4, $5, $6, $7, $8, $9, $10, - $11, $12, $13, 'cp=0.01'::TEXT, - 'max_surrogates=0', FALSE::BOOLEAN); + $11, $12, $13, NULL::TEXT, + NULL::TEXT, FALSE::BOOLEAN); $$ LANGUAGE SQL VOLATILE m4_ifdef(`__HAS_FUNCTION_PROPERTIES__', `MODIFIES SQL DATA', `'); @@ -1635,8 +1637,9 @@ CREATE OR REPLACE FUNCTION MADLIB_SCHEMA.tree_train( min_split INTEGER, min_bucket INTEGER ) RETURNS VOID AS $$ - SELECT MADLIB_SCHEMA.tree_train($1, $2, $3, $4, $5, $6, $7, $8, $9, $10, $11, $12, - 100::INTEGER, 'cp=0.01'::TEXT, 'max_surrogates=0', FALSE::BOOLEAN); + SELECT MADLIB_SCHEMA.tree_train($1, $2, $3, $4, $5, $6, $7, $8, $9, $10, + $11, $12, NULL::INTEGER, NULL::TEXT, + NULL::TEXT, FALSE::BOOLEAN); $$ LANGUAGE SQL VOLATILE m4_ifdef(`__HAS_FUNCTION_PROPERTIES__', `MODIFIES SQL DATA', `'); @@ -1654,8 +1657,8 @@ CREATE OR REPLACE FUNCTION MADLIB_SCHEMA.tree_train( min_split INTEGER ) RETURNS VOID AS $$ SELECT MADLIB_SCHEMA.tree_train($1, $2, $3, $4, $5, $6, $7, $8, $9, $10, $11, - ($11/3)::INTEGER, 100::INTEGER, 'cp=0.01'::TEXT, 'max_surrogates=0', - FALSE::BOOLEAN); + NULL::INTEGER, NULL::INTEGER, NULL::TEXT, + NULL::TEXT, FALSE::BOOLEAN); $$ LANGUAGE SQL VOLATILE m4_ifdef(`__HAS_FUNCTION_PROPERTIES__', `MODIFIES SQL DATA', `'); @@ -1672,8 +1675,8 @@ CREATE OR REPLACE FUNCTION MADLIB_SCHEMA.tree_train( max_depth INTEGER ) RETURNS VOID AS $$ SELECT MADLIB_SCHEMA.tree_train($1, $2, $3, $4, $5, $6, $7, $8, $9, $10, - 20::INTEGER, 6::INTEGER, 100::INTEGER, 'cp=0.01'::TEXT, - 'max_surrogates=0', FALSE::BOOLEAN); + NULL::INTEGER, NULL::INTEGER, NULL::INTEGER, + NULL::TEXT, NULL::TEXT, FALSE::BOOLEAN); $$ LANGUAGE SQL VOLATILE m4_ifdef(`__HAS_FUNCTION_PROPERTIES__', `MODIFIES SQL DATA', `'); @@ -1689,8 +1692,9 @@ CREATE OR REPLACE FUNCTION MADLIB_SCHEMA.tree_train( weights TEXT ) RETURNS VOID AS $$ SELECT MADLIB_SCHEMA.tree_train($1, $2, $3, $4, $5, $6, $7, $8, $9, - 10::INTEGER, 20::INTEGER, 6::INTEGER, 100::INTEGER, - 'cp=0.01'::TEXT, 'max_surrogates=0', FALSE::BOOLEAN); + NULL::INTEGER, NULL::INTEGER, NULL::INTEGER, + NULL::INTEGER, NULL::TEXT, NULL::TEXT, + FALSE::BOOLEAN); $$ LANGUAGE SQL VOLATILE m4_ifdef(`__HAS_FUNCTION_PROPERTIES__', `MODIFIES SQL DATA', `'); @@ -1705,8 +1709,9 @@ CREATE OR REPLACE FUNCTION MADLIB_SCHEMA.tree_train( grouping_cols TEXT ) RETURNS VOID AS $$ SELECT MADLIB_SCHEMA.tree_train($1, $2, $3, $4, $5, $6, $7, $8, - NULL::TEXT, 10::INTEGER, 20::INTEGER, 6::INTEGER, 100::INTEGER, - 'cp=0.01'::TEXT, 'max_surrogates=0', FALSE::BOOLEAN); + NULL::TEXT, NULL::INTEGER, NULL::INTEGER, + NULL::INTEGER, NULL::INTEGER, NULL::TEXT, NULL::TEXT, + FALSE::BOOLEAN); $$ LANGUAGE SQL VOLATILE m4_ifdef(`__HAS_FUNCTION_PROPERTIES__', `MODIFIES SQL DATA', `'); @@ -1720,9 +1725,9 @@ CREATE OR REPLACE FUNCTION MADLIB_SCHEMA.tree_train( split_criterion TEXT ) RETURNS VOID AS $$ SELECT MADLIB_SCHEMA.tree_train($1, $2, $3, $4, $5, $6, $7, - NULL::TEXT, NULL::TEXT, 10::INTEGER, 20::INTEGER, - 6::INTEGER, 100::INTEGER, 'cp=0.01'::TEXT, - 'max_surrogates=0', FALSE::BOOLEAN); + NULL::TEXT, NULL::TEXT, NULL::INTEGER, NULL::INTEGER, + NULL::INTEGER, NULL::INTEGER, NULL::TEXT, + NULL::TEXT, FALSE::BOOLEAN); $$ LANGUAGE SQL VOLATILE m4_ifdef(`__HAS_FUNCTION_PROPERTIES__', `MODIFIES SQL DATA', `'); @@ -1735,9 +1740,9 @@ CREATE OR REPLACE FUNCTION MADLIB_SCHEMA.tree_train( list_of_features_to_exclude TEXT ) RETURNS VOID AS $$ SELECT MADLIB_SCHEMA.tree_train($1, $2, $3, $4, $5, $6, - 'gini'::TEXT, NULL::TEXT, NULL::TEXT, 10::INTEGER, - 20::INTEGER, 6::INTEGER, 100::INTEGER, 'cp=0.01'::TEXT, - 'max_surrogates=0', FALSE::BOOLEAN); + NULL::TEXT, NULL::TEXT, NULL::TEXT, NULL::INTEGER, + NULL::INTEGER, NULL::INTEGER, NULL::INTEGER, NULL::TEXT, + NULL::TEXT, FALSE::BOOLEAN); $$ LANGUAGE SQL VOLATILE m4_ifdef(`__HAS_FUNCTION_PROPERTIES__', `MODIFIES SQL DATA', `'); @@ -1749,9 +1754,9 @@ CREATE OR REPLACE FUNCTION MADLIB_SCHEMA.tree_train( list_of_features TEXT ) RETURNS VOID AS $$ SELECT MADLIB_SCHEMA.tree_train($1, $2, $3, $4, $5, - NULL::TEXT, 'gini'::TEXT, NULL::TEXT, NULL::TEXT, - 10::INTEGER, 20::INTEGER, 6::INTEGER, 100::INTEGER, - 'cp=0.01'::TEXT, 'max_surrogates=0', FALSE::BOOLEAN); + NULL::TEXT, NULL::TEXT, NULL::TEXT, NULL::TEXT, + NULL::INTEGER, NULL::INTEGER, NULL::INTEGER, NULL::INTEGER, + NULL::TEXT, NULL::text, FALSE::BOOLEAN); $$ LANGUAGE SQL VOLATILE m4_ifdef(`__HAS_FUNCTION_PROPERTIES__', `MODIFIES SQL DATA', `'); -- ------------------------------------------------------------------------- http://git-wip-us.apache.org/repos/asf/incubator-madlib/blob/3eec0a82/src/ports/postgres/modules/recursive_partitioning/test/decision_tree.sql_in ---------------------------------------------------------------------- diff --git a/src/ports/postgres/modules/recursive_partitioning/test/decision_tree.sql_in b/src/ports/postgres/modules/recursive_partitioning/test/decision_tree.sql_in index 1863b64..28a4647 100644 --- a/src/ports/postgres/modules/recursive_partitioning/test/decision_tree.sql_in +++ b/src/ports/postgres/modules/recursive_partitioning/test/decision_tree.sql_in @@ -325,7 +325,7 @@ SELECT tree_train('dt_golf'::text, -- source table 'mse'::text, -- split criterion NULL::text, -- no grouping NULL::text, -- no weights - 10::integer, -- max depth + NULL::integer, -- max depth 6::integer, -- min split 2::integer, -- min bucket 8::integer, -- number of bins per continuous variable
