[MADLIB-1325] MLP does not stop even after tolerance reached

• Type: Bug
• Status: Closed
• Priority: Major
• Resolution: Fixed
• Fix Version/s: v1.16
• Component/s: Module: Neural Networks
• Assignee: Unassigned
• Reporter: Frank McQuillan
• Created: 12/Apr/19 21:43
• Updated: 29/May/19 23:59
• Resolved: 29/May/19 23:59

Description

madlib=# select madlib.mlp_classification(
madlib(# 'madlib_test_jl_train_dataset_for_pivotal', -- input table
madlib(# 'madlib_test_jl_mlp_training_model_20190312', -- output table
madlib(# 'ARRAY[g,h,i]',  -- independent var
madlib(# 'response',  -- dependent var
madlib(# ARRAY[10, 10, 10, 10], -- Number of neurons per layer, and specify number of hidden layers
madlib(# 'learning_rate_init = 0.01, lambda = 0, n_iterations = 10, tolerance = 0.0001', -- optimizer 
madlib(# 'relu', -- Activation function ('sigmoid' as default, 'tanh' and 'relu' available)
madlib(# NULL,  -- default weight (1)
madlib(# FALSE,  -- no warm start
madlib(# TRUE   -- not verbose
madlib(# );

INFO:  Iteration: 1, Loss: <1.42993215991>
INFO:  Iteration: 2, Loss: <1.21119290067>
INFO:  Iteration: 3, Loss: <1.21119290067>
INFO:  Iteration: 4, Loss: <1.21119290067>
INFO:  Iteration: 5, Loss: <1.21119290067>
INFO:  Iteration: 6, Loss: <1.21119290067>
INFO:  Iteration: 7, Loss: <1.21119290067>
INFO:  Iteration: 8, Loss: <1.21119290067>
INFO:  Iteration: 9, Loss: <1.21119290067>
 mlp_classification 
--------------------
 
(1 row)

Time: 26113.558 ms

This should have stopped after the tolerance was reached.