[
https://issues.apache.org/jira/browse/MADLIB-1209?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16373243#comment-16373243
]
Frank McQuillan commented on MADLIB-1209:
-----------------------------------------
looks better now
{code}
drop table if exists mlp_input;
create table mlp_input as
select 1 as fc_docid, 1 as responsive, array[1, 2, 3, 4, 5, 6, 7, 8, 9, 10] as
feature_vector
union all select 2, 0, array[2, 4, 6, 8, 10, 12, 14, 16, 18, 20];
DROP TABLE IF EXISTS mlp_model, mlp_model_summary, mlp_model_standardization;
SELECT madlib.mlp_classification(
'mlp_input', -- table containing the first training data set
'mlp_model', --output_table
'feature_vector', --independent_varname
'responsive', --dependent_varname
ARRAY[3,3], --hidden_layer_sizes
'learning_rate_init=0.003,n_iterations=10,tolerance=0', --optimizer_params
'relu', --activation
NULL, --weights
FALSE, --warm_start
TRUE --verbose
);
{code}
produces
{code}
INFO: Iteration: 2, Loss: <1.46611919957>
INFO: Iteration: 3, Loss: <1.45527321383>
INFO: Iteration: 4, Loss: <1.44470187518>
INFO: Iteration: 5, Loss: <1.43439130426>
INFO: Iteration: 6, Loss: <1.42432844068>
INFO: Iteration: 7, Loss: <1.41562036079>
INFO: Iteration: 8, Loss: <1.41267718821>
INFO: Iteration: 9, Loss: <1.40975892711>
mlp_classification
--------------------
(1 row)
{code}
> Neural nets fails on Greenplum 4.3.x when verbose argument set to TRUE
> ----------------------------------------------------------------------
>
> Key: MADLIB-1209
> URL: https://issues.apache.org/jira/browse/MADLIB-1209
> Project: Apache MADlib
> Issue Type: Bug
> Components: Module: Neural Networks
> Reporter: Frank McQuillan
> Priority: Minor
> Fix For: v1.14
>
>
> In
> {code}
> mlp_classification(
> source_table,
> output_table,
> independent_varname,
> dependent_varname,
> hidden_layer_sizes,
> optimizer_params,
> activation,
> weights,
> warm_start,
> verbose,
> grouping_col
> )
> {code}
> when the `verbose` argument is set to `TRUE` the following error results:
> {code}
> ERROR: plpy.SPIError: function array_length(double precision[], integer)
> does not exist (plpython.c:4656)
> LINE 4: _state_current[array_length(_sta...
> ^
> HINT: No function matches the given name and argument types. You may need to
> add explicit type casts.
> QUERY:
> SELECT
>
> _state_current[array_length(_state_current,1)] AS
> loss
> ,
> __madlib_temp_col_grp_key29688300_1519168141_31322353__
> FROM (
> SELECT grp_key AS
> __madlib_temp_col_grp_key29688300_1519168141_31322353__, state AS
> _state_current FROM madlib._gen_state($3, NULL, $4)
> ) subq1
>
> CONTEXT: Traceback (most recent call last):
> PL/Python function "mlp_classification", line 33, in <module>
> grouping_col
> PL/Python function "mlp_classification", line 290, in mlp
> PL/Python function "mlp_classification", line 455, in
> get_param_value_per_group
> PL/Python function "mlp_classification"
> {code}
> because `array_length` does not exist in Greenplum 4.3.x .
> It works OK in Greenplum 5.x and Postgres 9.x/10 however.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
