Hands-On Machine Learning for Algorithmic Trading

How to extract bottleneck features

Modern CNNs can take weeks to train on multiple GPUs on ImageNet. Fortunately, many researchers share their final weights. Keras, for example, contains pre-trained models for several of the reference architectures discussed previously, namely VGG16 and 19, ResNet50, InceptionV3 and InceptionResNetV2, MobileNet, DenseNet, NASNet, and MobileNetV2.

The bottleneck_features notebook illustrates how to download pre-trained VGG16 model, either with the final layers to generate predictions, or without the final layers, as illustrated in the following diagram, to extract the outputs produced by the bottleneck features:

Keras makes it very straightforward to download and use pre-trained models:

vgg19 = VGG19() ...