def test_bin_data(self):
binsizes = np.arange(4) + 1
shapes = (11, 4), (4, 11), (10, 3), (3, 10), (2, 2)
args = itertools.product(binsizes.tolist(), *zip(*shapes))
for nrows, ncols, binsize in args:
x = np.random.rand(nrows, ncols)
bins = cast(np.r_[:binsize:x.shape[0]], np.uint64)
binned = np.empty((bins.shape[0] - 1, x.shape[1]), np.uint64)
bin_data(x.view(np.uint8), bins, binned)
self.assertRaises(ValueError, bin_data, x, bins, binned)
self.assertRaises(ValueError, bin_data, x.view(np.int8), bins,
binned)
self.assertTupleEqual(binned.shape, (bins.size - 1, ncols))
self.assertEqual(binned.dtype, np.uint64)
def bin(self, cleared, binsize, reject_count=100, dropna=False):
"""Bin spike data by `binsize` millisecond bins.
Roughly, sum up the ones (and zeros) in the data using bins of size
`binsize`.
See :func:span.utils.utils.bin_data for the actual loop that
executes this binning. This method is a wrapper around that function.
Parameters
----------
cleared : array_like
The "refractory-period-cleared" array of booleans to bin.
binsize : numbers.Real
The size of the bins to use, in milliseconds
reject_count : numbers.Real, optional, default 100
Assign ``NaN`` to channels whose firing rates are less than this
number over the whole recording.
dropna : bool, optional
Whether to drop NaN'd values if any
Raises
------
AssertionError
* If `binsize` is not a positive number or if `reject_count` is
not a nonnegative number
Returns
-------
binned : SpikeGroupedDataFrame of float64
See Also
--------
span.utils.utils.bin_data
"""
assert binsize > 0 and isinstance(binsize, numbers.Real), \
'"binsize" must be a positive number'
assert reject_count >= 0 and isinstance(reject_count, numbers.Real), \
'"reject_count" must be a nonnegative real number'
ms_per_s = 1e3
bin_samples = cast(np.floor(binsize * self.fs / ms_per_s), np.uint64)
bins = np.arange(0, self.nsamples - 1, bin_samples, np.uint64)
shape = bins.shape[0] - 1, cleared.shape[1]
btmp = np.empty(shape, np.uint64)
bin_data(cleared.values.view(np.uint8), bins, btmp)
# make a datetime index of milliseconds
freq = binsize * datetools.Milli()
index = date_range(start=self.date, periods=btmp.shape[0],
freq=freq, name=r'$t\left(i\right)$',
tz='US/Eastern') + freq
binned = DataFrame(btmp, index=index, columns=cleared.columns,
dtype=np.float64)
# samples / (samples / s) == s
rec_len_s = self.nsamples / self.fs
# spikes / s
min_sp_per_s = reject_count / rec_len_s
# spikes / s * ms / ms == spikes / s
sp_per_s = binned.mean() * ms_per_s / binsize
# get rid of channels who have less then "reject_count" spikes over
# the whole recording
binned.ix[:, sp_per_s < min_sp_per_s] = np.nan
if dropna:
binned = binned.dropna(axis=1)
return SpikeGroupedDataFrame(binned)