Deep Learning with Keras and Tensorflow in R

A few words about CNNs - their structure and functionality - easy for anyone to understand.

Some links to detailed tutorials about CNNs, just in case you are interested.

What you need in order to build convolutional neural networks in R. How to install Python and its required packages.

YouTube tutorials on installing Python, keras and tensorflow on your computer.

A simple example on using CNNs for prediction with binomial response variable: predicting whether a bank customer will default or not. First of all, let's see how to prepare our data.

Time to set the parameters for our model and train it for predictions.

Let's measure the prediction accuracy of our new model.

Now we build a CNN for making predictions with a multinomial response variable: predict a wine quality (on three levels) using a few chemical and physical characteristics as predictors.

Let's train the network for our wine quality prediction.

Assessing the wine quality prediction accuracy in the test set.

How to convert our images into numbers that can be "crunched" by the R algorithms.

Decide which images will be used to train the model and which will be used to test it.

Let's put all together and create a convolutional neural network that tell humans from computers.

Time to assess how good is our network, by using is to distinguish the images in the test set.

Assess the prediction accuracy again, this time on brand new images that have not "seen" the model.

Preparing our photos for being included in the model as variables (i.e. convert them into large arrays of numbers).

Let's decide which photos will be used to train the model and assign them to the training set.

Let's create our model using a convolutional neural network.

Let's see if our model can recognize bears and foxes, using the photos in the test set as input variables for our new model.

Let's try to distinguish bears from foxes, this time using new photos.

Things get more complicated - we have three types of animals now. First, let's prepare our images for the analysis.

Dividing our sample of images into two sets: training set and test set.

Time to create our convolutional network and train it.

Tell bears, fox and mice apart, using the data in the test set.

Use new photos to assess the prediction accuracy of our model.

Now let's try to distinguish two special characters: asterisks and hashtags. First, we prepare our data for the analysis.

Divide our images into a training sample and a test sample.

Create and train our deep learning model.

Use the data in the test set to tell asterisks and hashtags apart.

In this lecture we explore the MNIST data set, which contains thousands of hand-written numbers.

Build our network using the images in the MNIST data set as input variables.

Use the images in the test set to recognize numbers.

Use our model on brand new data (hand-written numbers created by myself), to assess its prediction accuracy on "unseen" data.

Find a detailed description of all variables in the data sets here.

You can get the exercises here, in PDF format. If you need a description of each data set, please see the previous lecture in this section.

Download the data sets and source code for all lectures.