Etienne Ackermann
Etienne Ackermann
This has now been done, to some extent... ```python model_pos, model_speed = pos._kalman_smoother(Q=1, R=600) ``` where a good rule of thumb might be to set `R=100*pos.fs`
Still need to expose API a little better.
Update: this has been addressed for npl.plot(), but needs a more thorough treatment.
This can be a cool feature, and since then I have hooked up `PHMM.sample()`, but only for returning a feature matrix. To build / return a `BinnedSpikeTrainArray`, we need something...
By now I have increased confidence in the spatial information, but I also know that the sparsity is not being computed correctly... so this is being changed to a "bug"....
Example: ``` signal = nel.AnalogSignalArray([[1,2,3,5,10,11,12,15], [1,2,3,5,10,11,12,15]], fs=5) signal = signal.partition(n_epochs=5) signal[:,[0,1]] signal[[0,1],:] signal[[0,1,40],:] ``` What should expected behavior be? For consistency, probably a KeyError, unless a slice is used, right?
In fact, something like this definitely WAS happening in my `PoissonHMM` class, which is why I had that really tricky-to-track-down bug in my HMM code.
quick test of AnalogSignalArray suggests that this should not be an issue here, since the data inherently gets copied somewhere in the pipeline. Still, comprehensive and careful tests would be...
``` # Default keyword arguments ax = kwargs.get("ax", plt.gca()) ```
@sillyproject can we still compare your spatial information to mine that I implemented in nelpy? It would be useful to have verified results. Also, do you think you can implement...