Consider having Java streamer handle logging from Python code, to avoid asynchronous output

jpsember / java-ml

Java classes for machine learning

0 stars 0 forks source link

Closed jpsember closed 2 years ago

jpsember commented 2 years ago

At present, the Python outputs are 'separated in time' from corresponding Java ones; though this maybe only occurs on toy models

jpsember commented 2 years ago

[x] Have it log separate loss components: coord, size, objectness, classification
[x] It should have a bunch of fields which can optionally be added: epoch, train loss, test loss, and a flag for whether a tensor is included

jpsember commented 2 years ago

[x] Allow it to accept some strings to do formatting, e.g. 'subsection' with a bunch of blank lines and title