PythonLayer takes parameters by string #2871
Conversation
|
@philkr have you adapted this test to Python 3 already? |
shelhamer
force-pushed
the
python-layer-arg
branch
2 times, most recently
from
4130028 to
bd788bd
Compare
|
|
||
|
|
||
| def python_net_file(): | ||
| f = tempfile.NamedTemporaryFile(delete=False) |
There was a problem hiding this comment.
NamedTemporaryFile is created with mode='w+b' by default, which in python3 is a buffer, not a string. To fix this just call NamedTemporaryFile(mode='w+', delete=False).
|
The python3 error is easy to fix (see above). I have no idea why python2 fails on those warnings... |
|
@shelhamer @philkr Thank you so much for taking care of my previous PR! Other than this, I still have a bunch of PRs for python that are not finished. If I need to update and polish them, please let me know. |
|
According to this post in Stack Overflow, maybe we have to include |
|
Hmm, the branch CI had passed but this is an issue. I'll try to fix it shortly. |
shelhamer
force-pushed
the
python-layer-arg
branch
from
bd788bd to
9742bec
Compare
|
@jeffdonahue @longjon does anybody have a favorite name or is |
|
I think |
shelhamer
force-pushed
the
python-layer-arg
branch
4 times, most recently
from
245f920 to
39794ed
Compare
|
I changed the name of the property to |
| input: 'data' input_shape { dim: 10 dim: 9 dim: 8 } | ||
| layer { type: 'Python' name: 'mul10' bottom: 'data' top: 'mul10' | ||
| python_param { module: 'test_python_layer_with_param_str' | ||
| layer: 'SimpleParamLayer' param_str: "10" } } |
There was a problem hiding this comment.
Could we be consistent with single-quotes here?
|
Looks good except as noted. |
shelhamer
force-pushed
the
python-layer-arg
branch
from
39794ed to
a756cfe
Compare
|
Thanks for the glance @longjon. Fixed and merging. |
[pycaffe] PythonLayer takes parameters by string
Cherry-picking of the PythonLayer string parameter from #2001 by @tnarihi to give
PythonLayeraparam_strmember for passing parameters through thepython_paramproto message.This is a simple change in itself that leaves the choice of what and how to parameterize open.
There is no automatic
eval