[
https://issues.apache.org/jira/browse/MADLIB-1347?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16846181#comment-16846181
]
Frank McQuillan commented on MADLIB-1347:
-----------------------------------------
LGTM
{code}
madlib=# \d iris_model;
Table "public.iris_model"
Column | Type | Modifiers
------------+-------+-----------
model_data | bytea |
model_arch | json |
Distributed by: (model_data)
{code}
{code}
madlib=# select model_arch from iris_model;
-[ RECORD 1
]-------------------------------------------------------------------------------------------------------------------------------------------------------------------
model_arch |
| {"class_name": "Sequential", "keras_version": "2.1.6", "config":
[{"class_name": "Dense", "config": {"kernel_initializer": {"class_name":
"VarianceScaling", "config": {"distribution": "uniform", "scale": 1.0, "seed":
null, "mode": "fan_avg"}}, "name": "dense_1", "kernel_constraint": null,
"bias_regularizer": null, "bias_constraint": null, "dtype": "float32",
"activation": "relu", "trainable": true, "kernel_regularizer": null,
"bias_initializer": {"class_name": "Zeros", "config": {}}, "units": 10,
"batch_input_shape": [null, 4], "use_bias": true, "activity_regularizer":
null}}, {"class_name": "Dense", "config": {"kernel_initializer": {"class_name":
"VarianceScaling", "config": {"distribution": "uniform", "scale": 1.0, "seed":
null, "mode": "fan_avg"}}, "name": "dense_2", "kernel_constraint": null,
"bias_regularizer": null, "bias_constraint": null, "activation": "relu",
"trainable": true, "kernel_regularizer": null, "bias_initializer":
{"class_name": "Zeros", "config": {}}, "units": 10, "use_bias": true,
"activity_regularizer": null}}, {"class_name": "Dense", "config":
{"kernel_initializer": {"class_name": "VarianceScaling", "config":
{"distribution": "uniform", "scale": 1.0, "seed": null, "mode": "fan_avg"}},
"name": "dense_3", "kernel_constraint": null, "bias_regularizer": null,
"bias_constraint": null, "activation": "softmax", "trainable": true,
"kernel_regularizer": null, "bias_initializer": {"class_name": "Zeros",
"config": {}}, "units": 3, "use_bias": true, "activity_regularizer": null}}],
"backend": "tensorflow"}
{code}
> DL: Add model architecture column in model table
> ------------------------------------------------
>
> Key: MADLIB-1347
> URL: https://issues.apache.org/jira/browse/MADLIB-1347
> Project: Apache MADlib
> Issue Type: New Feature
> Components: Deep Learning
> Reporter: Ekta Khanna
> Priority: Major
> Fix For: v1.16
>
>
> Summary output table of fit writes out `model_arch_table` and `model_arch_id`
> columns which are used to find the model architecture to use in predict.
> Instead, we should also store the model architecture in the model table, so
> that the model is self sufficient during predict and is not dependent on
> another table.
> The reason for this feature is if the model_arch_table is dropped by
> accident, then all models trained so far will become unusable. A model must
> be self sustainable.
> Acceptance:
> 1. A column called `model_arch` must be created in `model` (output table of
> fit). It's ok to create it here and not the summary table.
> 2. Predict must use `model_arch` from `model` table, and not from the
> model_arch_table.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
