2martens/masterthesis
1
0
Fork 0
Code Issues Pull Requests Releases Wiki Activity

Removed reference to early stopping from docstring

Browse Source 
Signed-off-by: Jim Martens <github@2martens.de>
This commit is contained in:
Jim Martens
2019-04-04 17:25:34 +02:00
parent c83a65571b
commit 1f25122749
1 changed files with 0 additions and 10 deletions
Download Patch File Download Diff File Expand all files Collapse all files

10
src/twomartens/masterthesis/aae/train.py
View File

@ -71,16 +71,6 @@ def train_simple(dataset: tf.data.Dataset,
lr: initial learning rate (default: 0.002)
train_epoch: number of epochs to train (default: 80)
verbose: if True prints train progress info to console (default: True)
Notes:
The training stops early if for ``GRACE`` number of epochs the loss is not
decreasing. Specifically all individual losses are accounted for and any one
of those not decreasing triggers a ``strike``. If the total loss, which is
a sum of all individual losses, is also not decreasing and has a total
value of more than ``TOTAL_LOSS_GRACE_CAP``, the counter for the remaining grace period is
decreased. If in any epoch afterwards all losses are decreasing the grace
period is reset to ``GRACE``. Lastly the training loop will be stopped early
if the grace counter reaches ``0`` at the end of an epoch.
"""
# non-preserved tensors