search

kundajelab / dragonn

A toolkit to learn how to model and interpret regulatory sequence data using deep learning.
http://kundajelab.github.io/dragonn/
MIT License
254 stars 71 forks source link

problem with interpret #47

Open matan8 opened 6 years ago

matan8 commented 6 years ago

Hi, When I try to run the example interpretation I get an error (see below). Any idea? something wrong with my installation? train and predict seems to be fine.

getting deeplift scores... Traceback (most recent call last): File "/root/miniconda2/bin/dragonn", line 9, in load_entry_point('dragonn==0.1.3', 'console_scripts', 'dragonn')() File "build/bdist.linux-x86_64/egg/dragonn/main.py", line 203, in main File "build/bdist.linux-x86_64/egg/dragonn/main.py", line 164, in main_interpret File "build/bdist.linux-x86_64/egg/dragonn/models.py", line 186, in deeplift File "build/bdist.linux-x86_64/egg/deeplift/models.py", line 113, in get_target_contribs_func File "build/bdist.linux-x86_64/egg/deeplift/models.py", line 234, in _get_func File "build/bdist.linux-x86_64/egg/deeplift/models.py", line 49, in _get_func File "build/bdist.linux-x86_64/egg/deeplift/blobs/core.py", line 177, in update_mxts File "build/bdist.linux-x86_64/egg/deeplift/blobs/core.py", line 177, in update_mxts File "build/bdist.linux-x86_64/egg/deeplift/blobs/core.py", line 177, in update_mxts File "build/bdist.linux-x86_64/egg/deeplift/blobs/core.py", line 177, in update_mxts File "build/bdist.linux-x86_64/egg/deeplift/blobs/core.py", line 177, in update_mxts File "build/bdist.linux-x86_64/egg/deeplift/blobs/core.py", line 177, in update_mxts File "build/bdist.linux-x86_64/egg/deeplift/blobs/core.py", line 177, in update_mxts File "build/bdist.linux-x86_64/egg/deeplift/blobs/core.py", line 177, in update_mxts File "build/bdist.linux-x86_64/egg/deeplift/blobs/core.py", line 177, in update_mxts File "build/bdist.linux-x86_64/egg/deeplift/blobs/core.py", line 178, in update_mxts File "build/bdist.linux-x86_64/egg/deeplift/blobs/core.py", line 384, in _update_mxts_for_inputs File "build/bdist.linux-x86_64/egg/deeplift/blobs/convolution.py", line 668, in _get_mxts_increments_for_inputs File "build/bdist.linux-x86_64/egg/deeplift/blobs/convolution.py", line 631, in _get_input_grad_given_outgrad File "build/bdist.linux-x86_64/egg/deeplift/backend/theano_backend.py", line 236, in pool2d_grad AttributeError: 'Pool' object has no attribute 'grad'

jisraeli commented 6 years ago

@matan8 What version of theano are you using?

matan8 commented 6 years ago

Theano-1.0.1-py2.7.egg and Keras-0.3.3-py2.7.egg

jisraeli commented 6 years ago

I think the theano version is too recent for the version of deeplift used in dragonn. Can you try with theano 0.8 or 0.9?

matan8 commented 6 years ago

@jisraeli working without warnings now, thanks! now I encounter the next problem: I am trying to lower the positive threshold for interpretation but always get an error dragonn interpret: error: argument --pos-threshold: invalid int value: '5e-02' dragonn interpret: error: argument --pos-threshold: invalid int value: '0.05' trying to use just 5 doesn't give me the desired result.

jisraeli commented 6 years ago

Great. Does interpret command work for you with --pos-threshold 0?