Spike inference fails on traces with NaNs

[jzaremba]

As the title says.

Short-term solution would be to skip traces that have NaN's in them.

Long-term we'll want to be able to handle NaN's in the trace.

Additionally, need to ensure that sigma and gamma are estimated correctly with NaN values.

Example code that fails (modified from test_spikes.py):

import sima
import sima.spikes
import numpy as np
from scipy import signal
from scipy.stats import uniform, norm
import picos

RATE = 1  # mean firing rate of poisson spike train (Hz)
STEPS = 100  # number of time steps in data
TAU = 0.6  # time constant of calcium indicator (seconds)
DELTAT = 1/30  # time step duration (seconds)
sigma = 0.1    # standard deviation of gaussian noise
SEED = 2222    # random number generator seed

# Make a poisson spike trains
spikes = sima.spikes.get_poisson_spikes(
    deltat=DELTAT, rate=RATE, steps=STEPS, seed=SEED)

# Convolve with kernel to make calcium signal
np.random.seed(SEED)
gamma = 1 - (DELTAT / TAU)
CALCIUM = signal.lfilter([1], [1, -gamma], spikes)

# Make fluorescence traces with random gaussian noise and baseline
fluors = CALCIUM + norm.rvs(
    scale=sigma, size=STEPS) + uniform.rvs()
fluors[5:7] = np.nan

inference, fits, params = sima.spikes.spike_inference(fluors)