Closed decortja closed 3 years ago
At the time of the error (line 66 in base_imputer.py), {kwargs} = {'training': False}
if that helps.
Hi @decortja, thank you for spotting this!
I believe this is my mistake - the code for InductiveImputer
seems to be obsolete. Could you please check if PMI works with PseudoMaskImputer
? I refactored the code ~2 days ago. If this works, I'll remove InductiveImputer
from the repo to avoid future confusion. Many thanks again for reporting this issue!
Amazing! PseudoMaskImputer
working well. Was the patience set to 30 empirically?
Yes, it was. Feel free to use other values!
Sounds great. Also not deserving of a separate issue, but where is the output delivered to? I can't seem to find it. Particularly the per-gene imputation R^2 scores, like you have displayed in Figure 5.
We included a Jupyter notebook eval.ipynb
that loads the saved models and produces results. I hope this helps!
Hi all,
I have been able to use GAIN-GTEx on my data, but I am getting an error when I run the same data on PMI/the inductive imputer. Here are my settings:
The full error is pasted at the end, but the key error message is
ValueError: Layer model expects 4 input(s), but it received 5 input tensors.
As far as I can tell, the 5 inputs its receiving areFrom your paper, you say that the PMI imputer receives 4 inputs -- a table of gene data
x
, a mask for that datam
, numerical covariatesr
, and categorical covariatesq
. The imputer then generates a vectorb
, which operates onm
to generate the pseudomask. Would be great to get your input here on troubleshooting. Did the PMI imputer accidentally get passed 2 masks, or did it get passedb
along withm
? Is it supposed to usem
as input and calculateb
after?Full error message: