fmcquillan99 edited a comment on issue #402: DL: Enable warm start URL: https://github.com/apache/madlib/pull/402#issuecomment-498352780 (1) train for 3 iterations ``` DROP TABLE IF EXISTS iris_model, iris_model_summary; 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 3, -- num_iterations 0, -- GPUs per host 'iris_test_packed', -- validation dataset 1 -- metrics compute frequency ); SELECT * FROM iris_model_summary; -[ RECORD 1 ]-------------+-------------------------------------------------------------------------- source_table | iris_train_packed model | iris_model dependent_varname | class_text independent_varname | attributes model_arch_table | model_arch_library model_arch_id | 1 compile_params | loss='categorical_crossentropy', optimizer='adam', metrics=['accuracy'] fit_params | batch_size=5, epochs=3 num_iterations | 3 validation_table | iris_test_packed metrics_compute_frequency | 1 name | description | model_type | madlib_keras model_size | 0.7900390625 start_training_time | 2019-06-03 17:18:58.692221 end_training_time | 2019-06-03 17:19:04.165636 metrics_elapsed_time | {2.97064995765686,4.35413503646851,5.47341108322144} madlib_version | 1.16-dev num_classes | 3 class_values | {Iris-setosa,Iris-versicolor,Iris-virginica} dependent_vartype | character varying normalizing_const | 1 metrics_type | {accuracy} training_metrics_final | 0.833333313465 training_loss_final | 0.578077852726 training_metrics | {0.666666686534882,0.708333313465118,0.833333313465118} training_loss | {0.75943660736084,0.655552685260773,0.578077852725983} validation_metrics_final | 0.833333313465 validation_loss_final | 0.574034750462 validation_metrics | {0.666666686534882,0.666666686534882,0.833333313465118} validation_loss | {0.756515920162201,0.653064906597137,0.574034750461578} metrics_iters | {1,2,3} ``` (2) do warm start 2 times ``` 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 3, -- num_iterations 0, -- GPUs per host 'iris_test_packed', -- validation dataset 1, -- metrics compute frequency TRUE -- warm start ); ``` from orig fit ``` training_metrics_final | 0.833333313465 training_loss_final | 0.578077852726 training_metrics | {0.666666686534882,0.708333313465118,0.833333313465118} training_loss | {0.75943660736084,0.655552685260773,0.578077852725983} validation_metrics_final | 0.833333313465 validation_loss_final | 0.574034750462 validation_metrics | {0.666666686534882,0.666666686534882,0.833333313465118} validation_loss | {0.756515920162201,0.653064906597137,0.574034750461578} metrics_iters | {1,2,3} ``` warm start 1: ``` training_metrics_final | 0.891666650772 training_loss_final | 0.430346846581 training_metrics | {0.841666638851166,0.899999976158142,0.891666650772095} training_loss | {0.514156401157379,0.466687709093094,0.430346846580505} validation_metrics_final | 0.966666638851 validation_loss_final | 0.421454459429 validation_metrics | {0.866666674613953,0.966666638851166,0.966666638851166} validation_loss | {0.509902536869049,0.459847092628479,0.421454459428787} metrics_iters | {1,2,3} ``` warm start 2: ``` training_metrics_final | 0.941666662693 training_loss_final | 0.362128138542 training_metrics | {0.933333337306976,0.941666662693024,0.941666662693024} training_loss | {0.402873665094376,0.381914019584656,0.362128138542175} validation_metrics_final | 1 validation_loss_final | 0.345562160015 validation_metrics | {1,1,1} validation_loss | {0.391709893941879,0.367107719182968,0.345562160015106} metrics_iters | {1,2,3} ``` OK, looks like both metric and loss carry on from where they left off (3) warm start and change model_id ``` SELECT madlib.madlib_keras_fit('iris_train_packed', -- source table 'iris_model', -- model output table 'model_arch_library', -- model arch table 2, -- model arch id $$ loss='categorical_crossentropy', optimizer='adam', metrics=['accuracy'] $$, -- compile_params $$ batch_size=5, epochs=3 $$, -- fit_params 3, -- num_iterations 0, -- GPUs per host 'iris_test_packed', -- validation dataset 1, -- metrics compute frequency TRUE -- warm start ); training_metrics_final | 0.958333313465 training_loss_final | 0.313535839319 training_metrics | {0.958333313465118,0.958333313465118,0.958333313465118} training_loss | {0.345191985368729,0.328865617513657,0.313535839319229} validation_metrics_final | 1 validation_loss_final | 0.2869117558 validation_metrics | {1,1,1} validation_loss | {0.325278729200363,0.306510925292969,0.286911755800247} metrics_iters | {1,2,3} ``` OK, carries on from where it left off LGTM
---------------------------------------------------------------- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: [email protected] With regards, Apache Git Services
