lag CRP analysis crashing
Closed this issue · 4 comments
I tried running a lag CRP analysis on the "random" condition FRFR data:
import quail as q
data = q.load_egg('exp1.egg')
crp = data.analyze('lagcrp')
I was expecting crp to be a "fried egg" object that I could then plot using
crp.plot()
However, instead the code crashed with the following message (this is using a Docker image based on the Sherlock paper's Docker image, but with Quail upgraded to version 0.2.0):
---------------------------------------------------------------------------
AttributeError Traceback (most recent call last)
<ipython-input-1300-bd4500bf5146> in <module>()
----> 1 crp = fr_random_data.analyze('lagcrp')
/opt/conda/lib/python3.6/site-packages/quail/egg.py in analyze(self, analysis, **kwargs)
335 Calls analyze function
336 """
--> 337 return analyze(self, analysis=analysis, **kwargs)
338
339 class FriedEgg(object):
/opt/conda/lib/python3.6/site-packages/quail/analysis/analysis.py in analyze(egg, subjgroup, listgroup, subjname, listname, analysis, position, permute, n_perms, parallel, match, distance, features, ts)
151 opts.update({'ts' : ts})
152
--> 153 return FriedEgg(data=_analyze_chunk(egg, **opts), analysis=analysis,
154 list_length=egg.list_length, n_lists=egg.n_lists,
155 n_subjects=egg.n_subjects, position=position)
/opt/conda/lib/python3.6/site-packages/quail/analysis/analysis.py in _analyze_chunk(data, subjgroup, subjname, listgroup, listname, analysis, analysis_type, pass_features, features, parallel, **kwargs)
232 res = p.map(_analysis, chunks)
233 else:
--> 234 res = [_analysis(c) for c in chunks]
235
236 return pd.concat(res)
/opt/conda/lib/python3.6/site-packages/quail/analysis/analysis.py in <listcomp>(.0)
232 res = p.map(_analysis, chunks)
233 else:
--> 234 res = [_analysis(c) for c in chunks]
235
236 return pd.concat(res)
/opt/conda/lib/python3.6/site-packages/quail/analysis/analysis.py in _analysis(c)
211 else:
212 opts.update({'columns' : range(-data.list_length,data.list_length+1)})
--> 213 return pd.DataFrame([analysis(s, features=features, **kwargs)],
214 index=index, **opts)
215
/opt/conda/lib/python3.6/site-packages/quail/analysis/lagcrp.py in lagcrp_helper(egg, match, distance, ts, features)
118 if match is 'exact':
119 opts.update({'features' : 'item'})
--> 120 recmat = recall_matrix(egg, **opts)
121 if not ts:
122 ts = egg.pres.shape[1]
/opt/conda/lib/python3.6/site-packages/quail/analysis/recmat.py in recall_matrix(egg, match, distance, features)
47 if match=='exact':
48 features=['item']
---> 49 return _recmat_exact(egg.pres, egg.rec, features)
50 else:
51 return _recmat_smooth(egg.pres, egg.rec, features, distance, match)
/opt/conda/lib/python3.6/site-packages/quail/analysis/recmat.py in _recmat_exact(presented, recalled, features)
63 r = r_list.dropna().apply(get_feature).get_values()
64 r = np.vstack(list(filter(lambda x: x is not np.nan, r)))
---> 65 m = [np.where((p==x).all(axis=1))[0] for x in r]
66 result[li, :len(m)] = [x[0]+1 if len(x)>0 else np.nan for x in m]
67 return result
/opt/conda/lib/python3.6/site-packages/quail/analysis/recmat.py in <listcomp>(.0)
63 r = r_list.dropna().apply(get_feature).get_values()
64 r = np.vstack(list(filter(lambda x: x is not np.nan, r)))
---> 65 m = [np.where((p==x).all(axis=1))[0] for x in r]
66 result[li, :len(m)] = [x[0]+1 if len(x)>0 else np.nan for x in m]
67 return result
AttributeError: 'bool' object has no attribute 'all'
@jeremymanning this was a bug with handling a list for which there were 0 recalls. I fixed the bug on my fork for now (https://github.com/paxtonfitzpatrick/quail) but would rather look at that subject's data to see why there were zero recalls for the lists causing problems before merging. We should decide if this is a case we want quail to be able to handle or not. Throwing an error here might serve as a good indicator that something went wrong with the automatic transcription or during the experiment.
We should be able to handle cases where the participant didn't recall any words. That occasionally happens with undergraduate participants, and frequently happens in some patient populations-- so we definitely can't assume it's an error. So I'd say: submit a pull request!
done!