Using Machine Learning with TensorFlow 2.0 to solve a classification problem; The problem is to classify handwritten digits from a dataset of 70000 images. This dataset is a default in TF2-datasets, and it is known as MNIST dataset. MNIST Classification is considered the "Hello World!" of Machine Learning.
Classification problem. Targets are numbers from 0 to 9. Each image 2828 pixels. 784 pixels. It is on greyscale [0,255]. The approach for deep feed forward neural networks is to transform or “flatten” each image into a vector of length 7841. Our Input Layer has 784 Input Nodes.
We choose 2 hidden layers.
Output Layer: 10 nodes: 10 digits: 10 classes.
We use One-hot Encoding for the Outputs to be compared to Targets.
Activation Function = Softmax.
.
Train >> Validation >> Test
: This is actually what ML is about
- Prepare the data and preprocess it. Create training, validation and test datasets.
- Outline the model and choose the activation functions.
- Set the appropriate advance optimizers and loss function.
- Train the model.
- Test the accuracy of the model.
TF Datasets: TFDS: https://www.tensorflow.org/datasets/catalog/mnist.
TFDS has already split the dataset into “Training” and “Test” datasets.
Install TFDS Package using:
/pip install tensorflow-datasets
or
/conda install tensorflow-datasets