Make Model Configurable - Githubissues

janketj / usable_ml

GNU Affero General Public License v3.0

1 stars 0 forks source link

Make Model Configurable #3

Open lillijo opened 1 year ago

lillijo commented 1 year ago

Global Parameters @janketj

[ ] stop/ start training
[ ] send updates on the progress after every batch/epoch
[ ] send updates on accuracy/loss every x seconds/milliseconds
[ ] learning rate slider/field
[ ] batch size slider/field
[ ] test/train split slider/field
[ ] #epochs slider/field
[ ] seed field?
[ ] optimizer (stochastic gradient descent etc.) select
[ ] use gpu checkbox (cuda)
[ ] reset training
[ ] save current status
[ ] load current status

Model itself @ArpiHunanyan

[x] add layer: convolutional, linear, pooling ...
[x] remove layer
[x] change parameters: kernel size, padding, stride, activation function, # channels ...
[x] freeze/unfreeze layer
[x] layout of the layer (for discussion)
[x] create Block Class -> add_layer etc
[x] add fully connected layer/block
[x] which architecture to use for normalization and drop out
[x] how to connect blocks together: linked list -> call next.forward()
[x] predict correc w and h for next layer/block
[x] initial parameters : convblock automatically has conv layer with default params
[ ]

lillijo commented 1 year ago

Arpi -> model class
Jannik -> Model database
Dennis -> training class/function