René's URL Explorer Experiment

{"@context":"https://schema.org","@type":"DiscussionForumPosting","headline":"Loss function result differs from training Loss","articleBody":"Loss evaluated with a separate function:\r\n`let lossFn = mi.Func\u003csingle\u003e (loss) |\u003e arg2 input target`\r\n\r\ndiffers from the loss evaluated during training:\r\n`let trainFn () = Train.train trainable fullDataset trainCfg`\r\n\r\nWhen thisoption ist set:\r\n`SymTensor.Debug.DisableCombineIntoElementsOptimization \u003c- true`\r\nthe results are consistent again!\r\n\r\nTested via DeepPrivate:\r\n`Z:\\DeepPrivate\\****Models\\bin\\Release\\****Models.exe Z:\\DEVELOP\\DeepPrivate\\****Models\\cfgs\\COPYING\\LSTM\\80reluOutputTest\\Config.fsx`\r\n\r\nCode during training gets optimized differently?\r\n\r\nTested with [band 2277e5e]","author":{"url":"https://github.com/jklanger","@type":"Person","name":"jklanger"},"datePublished":"2017-03-29T14:12:12.000Z","interactionStatistic":{"@type":"InteractionCounter","interactionType":"https://schema.org/CommentAction","userInteractionCount":0},"url":"https://github.com/30/DeepNet/issues/30"}