Frank McQuillan created MADLIB-1363:
---------------------------------------
Summary: Reduce verbose output to console with fit()
Key: MADLIB-1363
URL: https://issues.apache.org/jira/browse/MADLIB-1363
Project: Apache MADlib
Issue Type: Improvement
Components: Deep Learning
Reporter: Frank McQuillan
Fix For: v1.16
fit() INFO and CONTEXT messages
(1) no validation_table, metrics_compute_frequency=0
{code}
SELECT madlib.madlib_keras_fit('iris_train_packed', -- source table
'iris_model', -- model output table
'model_arch_library', -- model arch table
1, -- model arch id
$$ loss='categorical_crossentropy',
optimizer='adam', metrics=['accuracy'] $$, -- compile_params
$$ batch_size=5, epochs=3 $$, -- fit_params
10 -- num_iterations
);
INFO: Processed 60 images: Fit took 0.567000865936 sec, Total was
0.757196903229 sec (seg0 slice1 10.128.0.41:40000 pid=13317)
CONTEXT: PL/Python function "fit_transition"
INFO: Processed 60 images: Fit took 0.55348110199 sec, Total was
0.741441011429 sec (seg1 slice1 10.128.0.41:40001 pid=13318)
CONTEXT: PL/Python function "fit_transition"
INFO: Time for training in iteration 1: 2.45737695694 sec
CONTEXT: PL/Python function "madlib_keras_fit"
{code}
change to
{code}
INFO: Time for training in iteration 1: 2.45737695694 sec
CONTEXT: PL/Python function "madlib_keras_fit"
{code}
(2) no validation_table, metrics_compute_frequency!=0
{code}
SELECT madlib.madlib_keras_fit('iris_train_packed', -- source table
'iris_model', -- model output table
'model_arch_library', -- model arch table
1, -- model arch id
$$ loss='categorical_crossentropy',
optimizer='adam', metrics=['accuracy'] $$, -- compile_params
$$ batch_size=5, epochs=3 $$, -- fit_params
10, -- num_iterations
0, -- gpus per host
NULL, -- validation table
1 -- metrics compute
frequency
);
INFO: Processed 60 images: Fit took 0.534310817719 sec, Total was
0.712550878525 sec (seg0 slice1 10.128.0.41:40000 pid=14501)
CONTEXT: PL/Python function "fit_transition"
INFO: Processed 60 images: Fit took 0.564456939697 sec, Total was
0.751413106918 sec (seg1 slice1 10.128.0.41:40001 pid=14502)
CONTEXT: PL/Python function "fit_transition"
INFO: Time for training in iteration 1: 2.28858995438 sec
CONTEXT: PL/Python function "madlib_keras_fit"
INFO: Time for evaluation in iteration 1: 0.188971996307 sec.
CONTEXT: PL/Python function "madlib_keras_fit"
INFO: Training set metric after iteration 1: 0.649999976158.
CONTEXT: PL/Python function "madlib_keras_fit"
INFO: Training set loss after iteration 1: 1.1202558279.
CONTEXT: PL/Python function "madlib_keras_fit"
{code}
change to
{code}
INFO: Time for training in iteration 1: 2.28858995438 sec
Time for evaluation in iteration 1: 0.188971996307 sec
Training set metric after iteration 1: 0.649999976158
Training set loss after iteration 1: 1.1202558279
CONTEXT: PL/Python function "madlib_keras_fit"
{code}
(3) yes validation_table, metrics_compute_frequency=0
SELECT madlib.madlib_keras_fit('iris_train_packed', -- source table
'iris_model', -- model output table
'model_arch_library', -- model arch table
1, -- model arch id
$$ loss='categorical_crossentropy',
optimizer='adam', metrics=['accuracy'] $$, -- compile_params
$$ batch_size=5, epochs=3 $$, -- fit_params
10, -- num_iterations
0, -- GPUs per host
'iris_test_packed' -- validation dataset
);
INFO: Processed 60 images: Fit took 0.552575826645 sec, Total was
0.734694004059 sec (seg0 slice1 10.128.0.41:40000 pid=18431)
CONTEXT: PL/Python function "fit_transition"
INFO: Processed 60 images: Fit took 0.549551010132 sec, Total was
0.734927892685 sec (seg1 slice1 10.128.0.41:40001 pid=18432)
CONTEXT: PL/Python function "fit_transition"
INFO: Time for training in iteration 1: 2.36340904236 sec
CONTEXT: PL/Python function "madlib_keras_fit"
{code}
change to
{code}
INFO: Time for training in iteration 1: 2.45737695694 sec
CONTEXT: PL/Python function "madlib_keras_fit"
{code}
(4) yes validation_table, metrics_compute_frequency=!0
{code}
SELECT madlib.madlib_keras_fit('iris_train_packed', -- source table
'iris_model', -- model output table
'model_arch_library', -- model arch table
1, -- model arch id
$$ loss='categorical_crossentropy',
optimizer='adam', metrics=['accuracy'] $$, -- compile_params
$$ batch_size=5, epochs=3 $$, -- fit_params
10, -- num_iterations
0, -- GPUs per host
'iris_test_packed', -- validation dataset
1 -- metrics compute
frequency
);
INFO: Processed 60 images: Fit took 0.57217502594 sec, Total was
0.817452907562 sec (seg0 slice1 10.128.0.41:40000 pid=19573)
CONTEXT: PL/Python function "fit_transition"
INFO: Processed 60 images: Fit took 0.554927110672 sec, Total was
0.800101041794 sec (seg1 slice1 10.128.0.41:40001 pid=19574)
CONTEXT: PL/Python function "fit_transition"
INFO: Time for training in iteration 1: 2.43148899078 sec
CONTEXT: PL/Python function "madlib_keras_fit"
INFO: Time for evaluation in iteration 1: 0.217161893845 sec.
CONTEXT: PL/Python function "madlib_keras_fit"
INFO: Training set metric after iteration 1: 0.524999976158.
CONTEXT: PL/Python function "madlib_keras_fit"
INFO: Training set loss after iteration 1: 0.984773635864.
CONTEXT: PL/Python function "madlib_keras_fit"
INFO: Time for evaluation in iteration 1: 0.205282926559 sec.
CONTEXT: PL/Python function "madlib_keras_fit"
INFO: Validation set metric after iteration 1: 0.600000023842.
CONTEXT: PL/Python function "madlib_keras_fit"
INFO: Validation set loss after iteration 1: 0.940379023552.
CONTEXT: PL/Python function "madlib_keras_fit"
{code}
change to
{code}
INFO: Time for training in iteration 1: 2.43148899078 sec
Time for evaluating training dataset in iteration 1: 0.217161893845 sec
Training set metric after iteration 1: 0.524999976158
Training set loss after iteration 1: 0.984773635864
Time for evaluating validation dataset in iteration 1: 0.205282926559 sec
Validation set metric after iteration 1: 0.600000023842
Validation set loss after iteration 1: 0.940379023552
CONTEXT: PL/Python function "madlib_keras_fit"
{code}
Note change in wording ^^^ because there are 2 evaluation times.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
