Github user mktal commented on the issue:
https://github.com/apache/incubator-madlib/pull/75
The SQL code used to obtain the above results (note that we assume two
tables mnist_train and mnist_test already exist in the database):
```sql
-- training on mnist_train and store the model in mnist_model
drop table if exists mnist_model;
create table mnist_model as
select
madlib.structure_svm_step(
ind_var::double precision[][],
dep_var::double precision[],
NULL::double precision[],
784::integer, -- number of features
10::integer, -- number of classes
8::integer, -- epoch
64::integer, -- batch_size
0.01::double precision, -- step size
0.001::double precision, -- regularization
1::double precision) as state
from mnist_train;
-- testing on mnist_test
select
1 - miss/total as accuracy
from (
select count(*)::float8 as miss
from mnist_test, mnist_model
where madlib.structure_svm_predict(state, ind_var) <> dep_var
) q, (select count(*)::float8 as total from mnist_test) q2
```
---
If your project is set up for it, you can reply to this email and have your
reply appear on GitHub as well. If your project does not have this feature
enabled and wishes so, or if the feature is enabled but not working, please
contact infrastructure at [email protected] or file a JIRA ticket
with INFRA.
---
