Lasagne

Lasagne is a lightweight library to build and train neural networks in Theano. It supports feed-forward networks such as Convolutional Neural Networks (CNNs), recurrent networks including Long Short-Term Memory (LSTM), and any combination thereof. Lasagne allows architectures of multiple inputs and multiple outputs, including auxiliary classifiers. It also offers many optimization methods including Nesterov momentum, RMSprop and ADAM.

Users can get a freely definable cost function and no need to derive gradients due to Theano's symbolic differentiation. It provides transparent support of CPUs and GPUs due to Theano's expression compiler. Lasagne grew out of a need to combine the flexibility of Theano with the availability of the right building blocks for training neural networks.

Its development is guided by a number of design goals which includes Simplicity, Transparency, Modularity, Pragmatism, Restraint and Focus. Lasagne is easy to use, easy to understand and easy to extend, to facilitate use in research. Its interface is kept small, with as few classes and methods as much as possible.

Every added abstraction and feature is always carefully scrutinized, to determine whether the added complexity is justified. Lasagne does not hide Theano behind abstractions, it directly process and return Theano expressions or Python / numpy data types.

Lasagne makes it easy to use components in isolation or in conjunction with other frameworks. It makes common use cases easy and does not overrate uncommon cases. Ideally, everything should be possible, but common use cases shouldn’t be made more difficult just to cater for exotic ones.