Brain Observatory notebook error - 'DataFrame' object has no attribute 'orientation'

Question

I am using the Allen Brain Observatory's notebook at: http://alleninstitute.github.io/AllenSDK/_static/examples/nb/brain_observatory.html

When I run cells 16 and 17, after running cell 17 I get this error:

AttributeError: 'DataFrame' object has no attribute 'orientation'

Here is the code of both cells:

from allensdk.brain_observatory.drifting_gratings import DriftingGratings

# example loading drifing grating data
#data_set = boc.get_ophys_experiment_data(512326618)
dg = DriftingGratings(data_set)
print("done analyzing drifting gratings")

import numpy as np
from matplotlib import pyplot as plt
%matplotlib inline

# filter for visually responding, selective cells
vis_cells = (dg.peak.ptest_dg < 0.05) &  (dg.peak.peak_dff_dg > 3)
osi_cells = vis_cells & (dg.peak.osi_dg > 0.5) & (dg.peak.osi_dg <= 1.5)
dsi_cells = vis_cells & (dg.peak.dsi_dg > 0.5) & (dg.peak.dsi_dg <= 1.5)

# 2-d tf vs. ori histogram
# tfval = 0 is used for the blank sweep, so we are ignoring it here
os = np.zeros((len(dg.orivals), len(dg.tfvals)-1))
ds = np.zeros((len(dg.orivals), len(dg.tfvals)-1))

for i,trial in dg.peak[osi_cells].iterrows():
    os[trial.ori_dg, trial.tf_dg-1] += 1

for i,trial in dg.peak[dsi_cells].iterrows():
    ds[trial.ori_dg, trial.tf_dg-1] += 1

max_count = max(os.max(), ds.max())

fig, (ax1, ax2) = plt.subplots(1,2)

# plot direction selectivity
im = ax1.imshow(ds, clim=[0,max_count], cmap='hot', interpolation='nearest')
ax1.set_xlabel('temporal frequency')
ax1.set_ylabel('direction')
ax1.set_xticks(np.arange(len(dg.tfvals)-1))
ax1.set_xticklabels(dg.tfvals[1:])
ax1.set_yticks(np.arange(len(dg.orivals)))
ax1.set_yticklabels(dg.orivals)
ax1.set_title('direction selective cells')

# plot orientation selectivity
im = ax2.imshow(os, clim=[0,max_count], cmap='hot', interpolation='nearest')
ax2.set_xlabel('temporal frequency')
ax2.set_ylabel('orientation')
ax2.set_xticks(np.arange(len(dg.tfvals)-1))
ax2.set_xticklabels(dg.tfvals[1:])
ax2.set_yticks(np.arange(len(dg.orivals)))
ax2.set_yticklabels(dg.orivals)
ax2.set_title('orientation selective cells')

# plot a colorbar
fig.subplots_adjust(right=0.9)
cbar_ax = fig.add_axes([0.95, 0.05, 0.05, 0.85])
cbar = fig.colorbar(im, cax=cbar_ax)
cbar.set_ticks(np.arange(0, max_count, 2)+0.5)
cbar.set_ticklabels(np.arange(0, max_count, 2, dtype=int))

plt.show()

The line pointed to in the Traceback is:

vis_cells = (dg.peak.ptest_dg < 0.05) &  (dg.peak.peak_dff_dg > 3)

Any help is appreciated. I wanted to use the 'allensdk' tag but I don't have 1500 reputation.

Answer 1

Can you confirm for me that you're using python3? There are a number of issues we're trying to address related to py3 (mostly h5py and PyTables) -- I ran into this one yesterday in fact.

The problem is that the stimulus table column names are read out of the NWB file with bytes ('b') strings in python 3, which are not compatible with pandas column indexing.

The fix is to patch this line:

https://github.com/AllenInstitute/AllenSDK/blob/master/allensdk/core/brain_observatory_nwb_data_set.py#L834

To decode the strings as utf-8. It looks like this:

features = [ f.decode('UTF-8') for f in features ]

I can roll this into the master branch if you like now. Otherwise there is another point release coming in June that will include this fix. I don't think we'll have full python3 support at that point yet, but we're working on it.