flowers.py is overtraining

weka511 / learn

Experiments in machine learning

GNU General Public License v3.0

0 stars 0 forks source link

Closed weka511 closed 3 years ago

weka511 commented 4 years ago

Over training is obvious after 10 epochs.

weka511 commented 4 years ago

Just tried Dopout(0.5) - a small improvement.

weka511 commented 4 years ago

regularizers.l2(0.001)

weka511 commented 4 years ago

kernel_regularizer=regularizers.l2(0.01)

weka511 commented 4 years ago

I've tried training with l1 and l2 filters, weight = 1e-3/1e-3. Accuracy plateaus at 0.8018 and 0.6703.

weka511 commented 4 years ago

Best run so far (uses more complex network and slower training rate). Which is responsible?

weka511 commented 3 years ago

Don't fix