All text snippets in german. The computer asks a number of questions through audio playback and the user answers with his voice. In the end, the computer plays a summary of what it understood.
Should be more or less the classic backpropagation approach (tf_nn_iris.py). This is heavily inspired from the example found on the Tensorflow site, but is extended with the adaption of the learning rate (coupled to network error), allowing for a significant reduction in neurons!
sklearn_iris_svm.py is an easy example for demonstrating how to use a SVM (Support Vector Machine) with python.
As of today, sklearns SVM is a wrapper for LibSVM. LibSVM is a binary tool with a nice command line interface (https://www.csie.ntu.edu.tw/~cjlin/libsvm/).
The iris data is the famous Fisher dataset: https://en.wikipedia.org/wiki/Iris_flower_data_set
Natively, vanilly SVMs only support binary classification. LibSVM however implements a Multiclass extension using modified equations. So LibSVMs Multiclass approach implements a logic that may not be the optimal strategy for your multiclass problem. Do you train one binary classifier per class? For each class, different features maybe decisive. Take into account classifying emails into folders: Some folders or tags are about persons, some about years, some about topics, some about importance... https://en.wikipedia.org/wiki/Ensemble_learning
Requirements: python 3 pip install sklearn