Edge MDT Custom Layers

1.1.1 verified Sat May 09 auth: no python

Edge MDT Custom Layers provides custom TensorFlow/Keras layers for edge machine learning deployment. Current version 1.1.1, requires Python >=3.10. Release cadence is irregular, primarily updates for bug fixes.

pip install edge-mdt-cl

Common errors

error ImportError: cannot import name 'MDTLayer' from 'edge_mdt_cl' ↓

cause Importing directly from top-level package instead of submodule.

fix

Use from edge_mdt_cl.layers import MDTLayer.

error AttributeError: module 'edge_mdt_cl' has no attribute 'layers' ↓

cause Using outdated version (<1.0.0).

fix

Upgrade to latest version: pip install --upgrade edge-mdt-cl.

Warnings

breaking Version 1.0.0 renamed all layers from `edge_mdt_cl.<name>` to submodule `edge_mdt_cl.layers`. ↓

fix Update imports: `from edge_mdt_cl.layers import LayerName` instead of `from edge_mdt_cl import LayerName`.

gotcha Requires TensorFlow 2.x; not compatible with TensorFlow 1.x. ↓

fix Install TensorFlow version >=2.0.

Imports

MDTLayer

wrong

from edge_mdt_cl import MDTLayer

correct

from edge_mdt_cl.layers import MDTLayer

MDTLayer is in the layers submodule.

EdgeConv2D

from edge_mdt_cl.layers import EdgeConv2D

Quickstart

Demonstrates creating an MDTLayer with 32 units and applying it to random input.

import tensorflow as tf
from edge_mdt_cl.layers import MDTLayer

layer = MDTLayer(units=32)
input_tensor = tf.random.normal([1, 64])
output = layer(input_tensor)
print(output.shape)