#Classification Deck
-
Testnotebook contains the Complete version of Basic IRIS data classification along with performance evaluation and algorithm comparision
-
Classification Simple consists of a shortened version of classifier with no evaluation
- IRIS data set consists of 50 samples from each of three species of Iris (Iris setosa, Iris virginica and Iris versicolor). Four features were measured from each sample: the length and the width of the sepals and petals, in centimetres. Based on the combination of these four features, Fisher developed a linear discriminant model to distinguish the species from each other.
-
MNIST Simple collects the required data itself when run for the first time and this project can be used to identify various handwritten digits and classify them into numerals between 0-9.
- MNIST uses about 47000 images to train a Deep Belief network powered by Tensorflow running on DSX.On a regular computer, The training takes about 20 min to complete but DSX does it under less than a minute.
- More Details on MNIST can be found here