Closed rob-p closed 9 years ago
That's great Rob - many thanks for the fix!
With regards to the recent improvements to Sailfish, how do see the development of Sailfish progressing, now that we also have Salmon? Are there situations where you would recommend using Sailfish rather than Salmon?
Hi Owen,
That's an excellent question, and one I'm still looking into myself. Currently, the updated version of Sailfish is somewhat faster than Salmon, but that speed gap will close substantially as soon as the RapMap-based changes to Salmon are merged into master. The way I see it (though we'll need to do more testing to confirm our intuition here), the most recent changes to Sailfish improve both its accuracy (it seems to perform very similarly to Kallisto across a wide range of tests) and speed. Because Salmon has an online inference phase, to learn auxiliary parameter models etc., it tends to come out on top in terms of accuracy. However, because of the CPU-cycles we burn during the online inference, there is a (probably unavoidable, but fairly moderate) speed hit. So, the "short answer" (which is still a bit tentative at this point but has increasing data to back it up) is that the new Sailfish will give you an accurate answer a bit faster than Salmon, but Salmon will tend to give a slightly more accurate answer. When I have a little bit of free time, I'm hoping to throw together a blog post explaining this in a bit more detail.
In addition to improvements in speed and accuracy, the interface of Sailfish has changed a bit with version >= 0.7.0. These changes should allow piquant to work with the new versions of Sailfish.