This version is a experimental section to try various features and interfaces of Flex.
We mainly focused to build flex-chain
in this version. The flex-chain
computes Bayesian neural network (a.k.a. probabilistic deep learning) at high speed for the data stream.
- Change the name of the project from Flip to Flex
2018-07-26
We mainly use computation resources flexibly by operating deepUpdate
selectively. At this time, the algorithm measures the statistical distance between the recent data stream and the the estimated PDF. This method improves throughput up to 4 times.
2018-03-26
We refine its usabilities and inner workings. Especially, the functions of Dist
including Sketch
doesn't consider the empty structure anuymore. Thus, many functions do not return options, making them easier to use.
- Many functions of
Dist
doesn't returnOption
#1 #3 - Implement
sample
for Sketch #25 - Support for-comprehension #26
- Fix minor bugs #16 #17 #20
2018-02-18
In this version, we have implemented the basic functionality of the Sketch
algorithm. Sketch
summarize real-valued random variable stream in order to estimate its probability density in a quick way, and to store and retrieve it only using the sublinear space.