search

majianjia / nnom

A higher-level Neural Network library for microcontrollers.
Apache License 2.0
815 stars 235 forks source link

Is v0.4.3 nnom.py support sigmoid? #183

Open songdaw opened 1 year ago

songdaw commented 1 year ago

Hi, My model use sigmoid activation like: activation = 'sigmoid' output = Activation(activation)(input) and I use generate_model() in nnom.py to convert model. I think should add ('activation' in layer.name and layer.get_config()['activation'] == 'sigmoid') to function is_shift_layer() .

majianjia commented 1 year ago

For sigmoid, please use Sigmoid layer instead of Activation("sigmoid") at the moment

songdaw commented 1 year ago

Sorry, I can't find how to use Sigmoid layer, but tf.keras.activations.sigmoid() which produce Lambda layer. Did I miss something? I use keras=2.6.0.

jonnor commented 3 months ago

Is this the right fix, just to add sigmoid to is_shift_layer ? Or are there other complications?

Btw, a workaround might be to use hard_sigmoid, it appears to be implemented

songdaw commented 3 months ago

sigmoid is already implemented, try replace act_hard_sigmoid() with act_sigmoid().

---- Replied Message ---- | From | Jon @.> | | Date | 03/30/2024 11:20 | | To | @.> | | Cc | @.>@.> | | Subject | Re: [majianjia/nnom] Is v0.4.3 nnom.py support sigmoid? (Issue #183) |

Is this the right fix, just to add sigmoid to is_shift_layer ? Or are there other complications?

Btw, a workaround might be to use hard_sigmoid, it appears to be implemented

— Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you authored the thread.Message ID: @.***>