def test_filterpeak():
"""Test finding the maximal point in a 3D filter"""
arr = np.zeros((5, 2, 2))
true_index = 7
arr.flat[true_index] = -1
true_indices = np.unravel_index(true_index, arr.shape)
idx, sidx, tidx = flt.filterpeak(arr)
assert true_index == idx
assert true_indices[0] == tidx
assert np.all(true_indices[1:] == sidx)
def test_filterpeak():
"""Test finding the maximal point in a 3D filter"""
arr = np.zeros((5, 2, 2))
true_index = 7
arr.flat[true_index] = -1
true_indices = np.unravel_index(true_index, arr.shape)
idx, sidx, tidx = flt.filterpeak(arr)
assert true_index == idx
assert true_indices[0] == tidx
assert np.all(true_indices[1:] == sidx)
def stimcut(data, expt, ci, width=11):
"""Cuts out a stimulus around the whitenoise receptive field"""
# get the white noise STA for this cell
wn = _loadexpt_h5(expt, 'whitenoise')
sta = np.array(wn[f'train/stas/cell{ci+1:02d}']).copy()
# find the peak of the sta
xc, yc = ft.filterpeak(sta)[1]
# cutout stimulus
X, y = data
Xc = ft.cutout(X, idx=(yc, xc), width=width)
yc = y[:, ci].reshape(-1, 1)
return Exptdata(Xc, yc)
def loadexpt(expt, cells, filename, train_or_test, history, nskip, cutout_width=None):
"""Loads an experiment from an h5 file on disk
Parameters
----------
expt : str
The date of the experiment to load in YY-MM-DD format (e.g. '15-10-07')
cells : list of ints
Indices of the cells to load from the experiment
filename : string
Name of the hdf5 file to load (e.g. 'whitenoise' or 'naturalscene')
train_or_test : string
The key in the hdf5 file to load ('train' or 'test')
history : int
Number of samples of history to include in the toeplitz stimulus
nskip : float, optional
Number of samples to skip at the beginning of each repeat
cutout_width : int, optional
If not None, cuts out the stimulus around the STA (assumes cells is a scalar)
"""
assert history > 0 and type(history) is int, "Temporal history must be a positive integer"
assert train_or_test in ('train', 'test'), "train_or_test must be 'train' or 'test'"
with notify('Loading {}ing data for {}/{}'.format(train_or_test, expt, filename)):
# get whitenoise STA for cutout stimulus
if cutout_width is not None:
assert len(cells) == 1, "cutout must be used with single cells"
wn = _loadexpt_h5(expt, 'whitenoise')
sta = np.array(wn[f'train/stas/cell{cells[0]+1:02d}']).copy()
py, px = ft.filterpeak(sta)[1]
# load the hdf5 file
with _loadexpt_h5(expt, filename) as f:
expt_length = f[train_or_test]['time'].size
# load the stimulus into memory as a numpy array, and z-score it
if cutout_width is None:
stim = zscore(np.array(f[train_or_test]['stimulus']).astype('float32'))
else:
stim = zscore(ft.cutout(np.array(f[train_or_test]['stimulus']), idx=(px, py), width=cutout_width).astype('float32'))
# apply clipping to remove the stimulus just after transitions
num_blocks = NUM_BLOCKS[expt] if train_or_test == 'train' and nskip > 0 else 1
valid_indices = np.arange(expt_length).reshape(num_blocks, -1)[:, nskip:].ravel()
# reshape into the Toeplitz matrix (nsamples, history, *stim_dims)
stim_reshaped = rolling_window(stim[valid_indices], history, time_axis=0)
# get the response for this cell (nsamples, ncells)
resp = np.array(f[train_or_test]['response/firing_rate_10ms'][cells]).T[valid_indices]
resp = resp[history:]
# get the spike history counts for this cell (nsamples, ncells)
binned = np.array(f[train_or_test]['response/binned'][cells]).T[valid_indices]
spk_hist = rolling_window(binned, history, time_axis=0)
return Exptdata(stim_reshaped, resp, spk_hist)
# In[6]:
spk = f['spikes/cell01']
# In[8]:
time = np.array(f['train/time'])
sta, tax = ft.getsta(time, whitenoise_train.X, spk, 35)
# In[11]:
Xcut = ft.cutout(whitenoise_train.X, idx=np.flipud(ft.filterpeak(sta)[1]), width=5)
# In[113]:
stc = np.zeros((35*11*11, 35*11*11))
for idx, s in enumerate(ft.getste(time, Xcut, spk, 35)):
sr = s.astype('float').ravel()
if sr.size == (35*11*11):
stc += np.outer(sr, sr)
if idx % 500 == 0:
print('{}'.format(100.*idx/len(spk)))
# In[ ]:
