We have now developed a good understanding of ANNs and their underlying technologies. I'd recommend libraries suchas TensorFlow and PyTorch for more complex architecture and for scaling up the training process on GPUs. However, you have a good headstart already. Most of the concepts discussed here are transferable to any other library. You will be using more or less the same activation functions and the same solvers, as well as most of the other hyperparameters discussed here. scikit-learn's implementation is still good for prototyping and for cases where we want to move beyond linear models without the need for too many hidden layers.
Furthermore, the solvers discussed here, such as gradient descent, are so ubiquitous in the field of machine learning, and so understanding their concepts is also helpful for understanding other algorithms that aren't neural networks. We saw earlier how gradient descent is used in training linear and logistic regressors...
