From 00460cb55550d944b09b77bd8b69fc85f4ee8099 Mon Sep 17 00:00:00 2001 From: =?UTF-8?q?Aur=C3=A9lien=20Geron?= Date: Wed, 13 Nov 2019 09:36:15 +0800 Subject: [PATCH] loss = metric * mean of sample weights, fixes #63 --- 12_custom_models_and_training_with_tensorflow.ipynb | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/12_custom_models_and_training_with_tensorflow.ipynb b/12_custom_models_and_training_with_tensorflow.ipynb index be59c78..e5bab44 100644 --- a/12_custom_models_and_training_with_tensorflow.ipynb +++ b/12_custom_models_and_training_with_tensorflow.ipynb @@ -1289,7 +1289,7 @@ "* the loss since the start of the epoch is the mean of all batch losses seen so far. Each batch loss is the sum of the weighted instance losses divided by the _batch size_ (not the sum of weights, so the batch loss is _not_ the weighted mean of the losses).\n", "* the metric since the start of the epoch is equal to the sum of weighted instance losses divided by sum of all weights seen so far. In other words, it is the weighted mean of all the instance losses. Not the same thing.\n", "\n", - "If you do the math, you will find that metric = loss * mean of sample weights (plus some floating point precision error)." + "If you do the math, you will find that loss = metric * mean of sample weights (plus some floating point precision error)." ] }, {