In the last of our 5-part mini series on Chatbots, we put our attention towards Theano – A specialty tech for Deep Learning.
What is Theano?
According to their website “Theano is a Python library that allows you to define, optimize, and evaluate mathematical expressions involving multi-dimensional arrays efficiently”. And by efficiently, they mean GPUs baby! This makes it highly suited to deep learning, because training deep learning models is essentially just linear algebra on very big matrices, over, and over, and over again.
What we love about Theano.
For starters, it allows us to transparently use the power of the GPU, speeding up training of machine learning models. Being in Python, it’s far easier to use than it’s C++, C and FORTRAN (I kid you not) counterparts. Another awesome thing about Theano is that it’s open source! That means we don’t need to pay to train models with it. But more importantly, it means that a whole ton more eyes have scrutinized and contributed to it.
On that note: it’s a mature software project and has been used in large-scale scientific investigations since 2007. This means the community is active, well developed and knowledgeable. It’s well-supported… there is an existing library and tutorial to build any deep learning capability with Theano. Due to its maturity, Theano is used as a backend to Keras (arguably the most wonderfully abstracted machine learning library).
We love its ease of use. We love that it is a Python-based product and we love the wealth of support we can access, because of the nature of the product.
Where have we used this software?
We use Theano to train deep learning systems for our clients. In particular, we’ve used libraries built with Theano in programming long and short-term memory neural networks for use in natural language processing tasks, enabling a machine to “understand” the context of words we used.
For more on Theano, check out this website. Also catch some of our talented Software engineers talking about our experiences with Bots at Botcon 2017, happening on the 25th of May in Cape Town, South Africa. This concludes the 5-part mini-series. If you missed the previous post, you can go back and view all items here.