My first experiments with Machine Learning (TensorFlow in Jupyter notebook).
-
My first project uses a Jupyter notebook and a list of a quarter of a million credit card transaction to make a neural network that detects the fraudulent ones.
-
My second project classifies 131 species of fruit (and vegetables and nuts). So here the I use a convolutional neural network that takes images as input.
-
My third project tries to predict the
sin(x)
fromx
, a network even simpler than for the credit card transactions. However, the real goal is to run the model on the ESP32, a goal that is reached. The same sketch even runs on ESP8266, albeit slower. There are two intermezzos. The first maps a piece-wise linear curve to a two-layer neural net. The second uses that technique to hand-craft a sine predictor. -
The fourth project is no longer to educate me. This implements rock-paper-scissors on an ESP32 with a camera connected. A supporting project (ESP32CAM) captured the images that were use for training the neural network. A follow-up project (TFLcam) is the resulting end-product. It can be used as a sensor in a rock-paper-scissors Lego Mindstorms robot.
(end of doc)