I want to build a neural network that implements an approximation to a non-linear sensor behavior shown in the table. It has one input and one output. The hidden layer should have a maximum of five neurons. Is it possible to train such a network in Edge Impulse for later implementation on a Arduino ESP32 for example?
|Temperature|Voltage|
|-25|4.652|
|-20|4.542|
|-10|4.244|
|0|3.836|
|10|3.333|
|20|2.778|
|25|2.5|
|30|2.227|
|40|1.731|
|50|1.32|
|60|0.989|
|70|0.74|
|80|0.554|
|90|0.417|
|100|0.316|
|110|0.241|
|120|0.186|
|125|0.164|
Hi @ckuehnel,
It looks like you want to do regression. You will want to upload your data with the ground-truth output value as your label. Please follow this regression guide for more information.
From there, you can either define your network architecture in the neural network settings page:
Or you can switch to Expert mode if you are more comfortable writing Keras code.