TensorFlow is an Open Source Software Library for Machine Intelligence
What is a Data Flow Graph?
Data flow graphs describe mathematical computation with a directed graph of nodes & edges. Nodes typically implement mathematical operations, but can also represent endpoints to feed in data, push out results, or read/write persistent variables. Edges describe the input/output relationships between nodes. These data edges carry dynamically-sized multidimensional data arrays, or tensors. The flow of tensors through the graph is where TensorFlow gets its name. Nodes are assigned to computational devices and execute asynchronously and in parallel once all the tensors on their incoming edges becomes available.
I was playing with it just the other day It’s really awesome to develop intuitions about the effects of hyperparameters (learning rate, L1/L2 regularization settings, layer size, number of layers, etc) and about how the features come into existence at the different depths of the network.
It’s always good to have options. As for me, I love Torch, but it seems Theano has always been “bigger” (considering how much new research used it, for example) and Google now has explicit directions about going with Tensorflow, so that support is now lost (well, there’s still Facebook and Twitter afaik)
I’ve been studying much about reinforcement learning recently (to develop a trading strategy for polo haha) and I noticed that maybe some of the recent advances (async algorithms, specifically: arXiv:1602.01783) would fit into the SAFE network’s structure pretty well, because they require less real-time communication.
About the trading thing? Well the idea is that it should be possible to use reinforcement learning to train a network to turn recent market data into a policy (trading decisions) that is profitable on the long run. I’m trying to do it on a portfolio level, so multiple agents would use the same rules (network) and the reward is computed on the cumulative result. The good thing about RI is that it can be model-free, which is analogous to the basic tinkering we already know works in real life.
NOTE: If not about the trading thing, but about the RI thing in general, then I have already added a link to a research paper But here’s a YouTube video from the author of that paper from before it was published: