ESP
ESP copied to clipboard
damellis
The Example-based Sensor Predictions (ESP) system applies machine learning to real-time sensor data.
In addition to existing one. The minimal one would include just pipeline and input/output stream and tunable parameters, but not heuristics. Maybe no calibration either? Or hard-coded calibration?
And keep the title as "1", "2", "3", etc.
Should we reduce to 5 or 6 classes? Or maybe show only a few by default but have a way to "add" more? Maybe still with a limit of 9.
@nebgnahz do you have time to do this today? (It's probably not as important as on Mac and Windows, since the compilation pre-requisites aren't as big / slow to download...
For instance, it would be good to show a cursor under the mouse location. Also, it can be hard to see the separation between the different plots. And it might...
For instance, for classifier that support null rejection, we could automatically generate the "variability" tuneable parameter.
It would be good to be able to restart the program receiving predictions from ESP using the TCP output stream and have the communication resume. This could be done either...
Maybe using the Timer1 library (https://www.pjrc.com/teensy/td_libs_TimerOne.html, https://www.arduino.cc/en/Reference/CurieTimerOne)?
Can we get the template length / number of live data points that were used to match against the predicated class and highlight all of them in the live plot?...
This should probably be actual axis labels (as opposed to showing ranges in the info text).