extract epoch based signal

[lungsi]

Added two methods within the class BaseSignal these are: (staticmethod) _rescale_epoch_time and extract_for_epoch. Therefore when a neo object signal is created (which in turn is an instance of the child class of BaseSignal) it has the method extract_for_epoch. Passing the epochs (created neo.Epoch) into this method will return a list of extracted signals (specific to the epochs).

>>> import neo
>>> import quantities as pq
>>> import numpy as np
>>> sigarr = neo.AnalogSignal([[1], [2], [3], [4], [5], [6]], units='mV',
                              sampling_rate=1*pq.Hz)
>>> epc = neo.Epoch(times=np.array([0, 3000])*pq.ms, durations=[1000, 2000]*pq.ms,
                    labels=np.array(['btn0', 'btn1'], dtype='S'))
>>> epochsignals = sigarr.extract_for_epoch(epc)
>>> epochsignals
[<AnalogSignal(array([[1]]) * mV, [0.0 s, 1.0 s], sampling rate: 1.0 Hz)>,
 <AnalogSignal(array([], shape=(0, 1), dtype=int64) * mV, [3.0 s, 3.0 s],
 sampling rate: 1.0 Hz)>]
>>> len(epochsignals)
2
>>> epochsignals[0]
<AnalogSignal(array([[1]]) * mV, [0.0 s, 1.0 s], sampling rate: 1.0 Hz)>

[pep8speaks]

Hello @lungsi! Thanks for updating the PR.

• In the file neo/core/basesignal.py, following are the PEP8 issues :

Line 13:69: W291 trailing whitespace
Line 40:78: W291 trailing whitespace
Line 42:73: W291 trailing whitespace
Line 61:68: W291 trailing whitespace
Line 63:58: W291 trailing whitespace
Line 86:61: W291 trailing whitespace
Line 311:1: W293 blank line contains whitespace
Line 364:28: E225 missing whitespace around operator

Comment last updated on April 19, 2018 at 11:49 Hours UTC

[coveralls]

Coverage decreased (-0.02%) to 48.87% when pulling b2f075c on lungsi:epochextract into db9357c on NeuralEnsemble:master.

[coveralls]

Coverage decreased (-0.02%) to 48.87% when pulling 7ee2f82 on lungsi:epochextract into db9357c on NeuralEnsemble:master.

[JuliaSprenger]

typo: this should be 'respective'

[JuliaSprenger]

Hi Lungsi!
Thanks for your contribution.
The utility function you are proposing is indeed part of a collection of functions we are missing for quite some time. However, I am not sure if this type of utility functions should be a method of the BaseSignal class or should go into a separate neo_utility function collection.
@apdavison @samuelgarcia I think we discussed this already once, based on the neo_utils collection here (https://web.gin.g-node.org/INT/multielectrode_grasp/src/master/code/neo_utils.py), which we planned to include at some point. What is your current opinion on the topic?

If this is going to stay a BaseSignal method, I would suggest to include the code of the _rescale_epoch_time directly in the extract_for_epoch, since it does not make sense to have an epoch specific function in the BaseSignal class.

[JuliaSprenger]

maybe cut_by_epoch or slice_by_epoch is more telling?

[lungsi]

I have not particular preference, I used the verb extract because it returns the cut/sliced signal.

[lungsi]

Hi Julia,
Thanks for the comments.
The point regarding the function performing a utility is a good one. I started out thinking that way but since there were no current utility.py script/source fille so I took the current approach which is to make signal have the method where one just had to pass the epoch. If there's going to be a script containing the utility methods I thinks its ok to have it there.

With regards to why I have _rescale_epoch_time and extract_for_epoch, not just one is because it avoids code duplication and also one has only 1-reason to change/edit either methods.

[apdavison]

I prefer slice_by_epoch to match time_slice

[apdavison]

I agree with @JuliaSprenger on moving _rescale_epoch_time inside the function where it is used. This does not imply code duplication, since you can define a function inside a function.

Alternatively, it could be moved out of the BaseSignal definition into a utility function, as checking two Neo objects have the same units must surely be a common need.

[apdavison]

I believe this is an implementation of #387 ?

[apdavison]

shouldn't this be "!=" rather than "is not"?

[samuelgarcia]

Dear all.
I made some coments about slicing here #526

I have implemented proxy objects here #461.
This is not merged yet, but it is almost ready (except some side effect errors and pep87).

For doing this I would prefer to write external function as proposed by @JuliaSprenger and map them to method as proposed by @apdavison.

So with proxy objects the external function could be hacked easily to work with standard/proxy objects.

My proposal is AnalogSignal.extract_for_epoch should call external neo.io.utils_slice.slice_sig_by_epoch (or whatever)

External function will also help for global slicing at segment/block level.

EDIT: I am totally repeating what @JuliaSprenger and @apdavison said before sorry.

[apdavison]

I'm postponing this to the next milestone due to lack of time to work on it.

[apdavison]

closing now as this PR is very old, but we will revisit the idea