Intro-to-Tensorflow-2.0.0

Introductory Projects to Tensorflow 2.x and Data Preprocessing in Python.

Projects

Projects presented in increasing complexity

• Hello Tensorflow: Simple single-layer neural network that predicts the y value of a linear equation (y = 2x - 1).
  • Single dense layer with single neuron.
  • Makes use of the mean squared error loss function and the stochastic gradient descent optimizer.
• Celcius to Farenheit: Simple neural network that is able to convert celcius degrees to Farenheit. (Not that this can't be achieved by pen and paper but where's the fun in that!? :) ).
  • Single dense layer with single neuron.
  • Makes use of the mean squared error loss function and the adam optimizer.
  • Uses early stopping by implementing Tensorflow's on_epoch_end() function so as to prevent overfitting.
• Fashion MNIST: A simple Convolutional Neural Network that predicts the type of clothing artifact from a dataset of 10 different classes.
  • Implements Keras' Conv2D and MaxPooling2D layers so as increase feature detection accuracy by the model while decreasing input to be processed by the model's dense layers.
  • Uses early stopping by implementing Tensorflow's on_epoch_end() function so as to prevent overfitting.
  • Uses Keras' sparse_categorical_crossentropy loss function to handle multi-class classification.
• Cats vs Dogs: CNN trained on over 2,000 images to predict whether image contains a cat or a dog.
  • Uses early stopping by implementing Tensorflow's on_epoch_end() function so as to prevent overfitting.
  • Makes use of Tensorflow's ImageDataGenerator so as to simplify image data preprocessing.
  • Implements image augmentation on the test dataset so as ti increase the variation of image styles that the model is trained on.
  • Implements the Keras Dropout layer so as to prevent overfitting while potentially increasing both training and validation accuracies.
  • Uses the RMSprop optimizer so as to have maximum control over the model's learning rate.

Happy Hacking,

Gerry.