Python Tutorial: Image classification with Keras

Want to learn more? Take the full course at https://learn.datacamp.com/courses/image-processing-with-keras-in-python at your own pace. More than a video, you'll learn hands-on coding & quickly apply skills to your daily work. --- Let's use Keras to classify images. We'll start by using a fully connected network like the ones that you saw in the deep learning course. We start by importing the Sequential model and initializing it. To construct our network we will use densely connected layers. Every unit in every layer is connected to all the units in the previous layer. The first layer of the network is connected to all the pixels in the input image. The training data in this case are images of clothes: 50 samples, each of 28 by 28 pixels. The last dimension has length 1, because the images are black and white. 50 is a rather small number of training samples, but we'll use that here for simplicity, and so that training proceeds rapidly. The first layer is a set of densely connected units. We'll use 10 units here, but you could use another number - more units would increase the complexity of the network and its capacity to represent complex inputs. To facilitate learning, we're using a rectified linear unit or "relu" as the activation. This should be familiar from the deep learning course. The input_shape keyword argument tells us how many inputs each of these units should expect. In this case, it is 784: one connection from each one of the pixels in the image (28 by 28 is 784). We add another hidden layer, also with 10 units. Also with the "relu" activation. The output of our network is a fully connected layer with a unit for each class of inputs: 3 classes for the three types of clothing. The output unit uses a "softmax" activation to decide which of the three classes was presented. This diagram shows the network and all its connections. Next, we compile the model. We choose the optimizer to use (adam), and a loss function: categorical cross-entropy, which is