I converted a TF model to ONNX and then ONNX model to Caffe2. The conversion happened successfully. However, I am getting a RunTime Error when trying to load and infer from the obtained model.
This is the error that I am receiving. How do I add the attribute 'is_true' to SpatialBN node?
I went through the pytorch repo and saw this issue, however, it is unresolved.
In the code base of ONNX here, it adds is_test attribute for opset >=7 and I am using 8. However, it is still giving the error.
[W common_gpu.cc:35] Insufficient cuda driver. Cannot use cuda.
[W init.h:137] Caffe2 GlobalInit should be run before any other API calls.
[W init.h:137] Caffe2 GlobalInit should be run before any other API calls.
[W predictor_config.cc:90] Caffe2 is compiled without optimization passes.
[E operator_schema.cc:101] Argument 'is_test' is required for Operator 'SpatialBN'.
Traceback (most recent call last):
File "main.py", line 91, in <module>
test_caffe("mod-caffe-net.pb", "mod-caffe-init-net.pb", "../data/mouth")
File "main.py", line 70, in test_caffe
predictor = workspace.Predictor(param_values, model_net)
File "/home/ubuntu/.local/lib/python3.6/site-packages/caffe2/python/workspace.py", line 187, in Predictor
return C.Predictor(StringifyProto(init_net), StringifyProto(predict_net))
RuntimeError: [enforce fail at operator.cc:199] schema->Verify(operator_def). Operator def did not pass schema checking: input: "conv1/Relu:0" input: "batchNorm1/gamma/read/_1__cf__1:0" input: "batchNorm1/beta/read/_0__cf__0:0" input: "batchNorm2/moving_mean/read/_6__cf__6:0" input: "batchNorm1/moving_variance/read/_3__cf__3:0" output: "batchNorm1/FusedBatchNorm:0" name: "batchNorm1/FusedBatchNorm" type: "SpatialBN" arg { name: "epsilon" f: 0.001 } device_option { device_type: 0 device_id: 0 }
The issue is resolved. I was using the command-line utility suggested on their README. However, it points to their tutorial in deprecated version of the code.
The command-line utility (installed using the pip install onnx-caffe2) still has the _known_opset_version = 3. This was causing the error. After I used the conversion utility through Python APIs in PyTorch library by importing,
from caffe2.python.onnx.backend import Caffe2Backend as c2
I was successfully able to run inference on the converted model.
Related
Reproduction Steps
Setup environment
python3.10 -m venv venv
source activate venv/bin/activate
pip install -U pip
pip install tensorflow==2.8.0 tf2onnx==1.13.0 packaging==23.0
Create a model with TextVectorizer
import tensorflow as tf
text_dataset = tf.data.Dataset.from_tensor_slices(["foo", "bar", "baz"])
vectorize_layer = tf.keras.layers.TextVectorization(max_tokens=100, output_mode='int', output_sequence_length=4)
vectorize_layer.adapt(text_dataset)
model = tf.keras.models.Sequential([
tf.keras.Input(shape=(1,), dtype=tf.string),
vectorize_layer
])
model.save("text_vectorizer")
Try to convert the model to ONNX
python -m tf2onnx.convert --saved-model text_vectorizer --output text_vectorizer.onnx
Result:
<...>
/tf2onnx/lib/python3.10/site-packages/tf2onnx/tf_utils.py:58: FutureWarning: In the future `np.str` will be defined as the corresponding NumPy scalar.
np_data = np_data.astype(np.str).astype(object)
Traceback (most recent call last):
File "/tf2onnx/lib/python3.10/site-packages/tf2onnx/tf_utils.py", line 58, in tf_to_onnx_tensor
np_data = np_data.astype(np.str).astype(object)
File "/tf2onnx/lib/python3.10/site-packages/numpy/__init__.py", line 305, in __getattr__
raise AttributeError(__former_attrs__[attr])
AttributeError: module 'numpy' has no attribute 'str'.
`np.str` was a deprecated alias for the builtin `str`. To avoid this error in existing code, use `str` by itself. Doing this will not modify any behavior and is safe. If you specifically wanted the numpy scalar type, use `np.str_` here.
The aliases was originally deprecated in NumPy 1.20; for more details and guidance see the original release note at:
https://numpy.org/devdocs/release/1.20.0-notes.html#deprecations. Did you mean: 'std'?
During handling of the above exception, another exception occurred:
<...>
File "/tf2onnx/lib/python3.10/site-packages/tf2onnx/tf_utils.py", line 63, in tf_to_onnx_tensor
raise RuntimeError("Not support type: {}".format(type(np_data.flat[0])))
RuntimeError: Not support type: <class 'bytes'>
How to fix the problem? Is there a way to make this conversion?
Below is the line of the code that pykalman example is using to load the matlab file robot.mat. However, it is throwing an error with message ValueError: Unknown mat file type, version 100, 110.
I am new to reading matlab files in Python. Can anyone help me understand exactly what version it is expecting and how to get around the issue?
data = io.loadmat(join(module_path, 'data', 'robot.mat'))
Traceback (most recent call last):
File "C:\ProgramData\Anaconda2\lib\site-
packages\IPython\core\interactiveshell.py", line 2881, in run_code
exec(code_obj, self.user_global_ns, self.user_ns)
File "<ipython-input-4-0439d709929f>", line 1, in <module>
data = load_robot()
File "C:\ProgramData\Anaconda2\lib\site-packages\pykalman\datasets\base.py",
line 45, in load_robot
data = io.loadmat(join(module_path, 'data', 'robot.mat'))
File "C:\ProgramData\Anaconda2\lib\site-packages\scipy\io\matlab\mio.py",
line 135, in loadmat
MR = mat_reader_factory(file_name, appendmat, **kwargs)
File "C:\ProgramData\Anaconda2\lib\site-packages\scipy\io\matlab\mio.py",
line 59, in mat_reader_factory
mjv, mnv = get_matfile_version(byte_stream)
File "C:\ProgramData\Anaconda2\lib\site-
packages\scipy\io\matlab\miobase.py",
line 241, in get_matfile_version
raise ValueError('Unknown mat file type, version %s, %s' % ret)
ValueError: Unknown mat file type, version 100, 110
I'm guessing here (and adding one more idea to hpaulj's good overview):
First i checked the file manually and opening in an editor reads as:
MATLAB 5.0 MAT-file, Platform: MACI64, Created on: Tue Jun 26 00:59:04 2012
...
Therefore there should be no problem in regards to scipy (if the mat-file is actually following this "header").
Now it seems there are some open issues with that project with the most relevant beeing:
Unable to load robot.mat #59
(Don't be surprised too much with the different eror-values as i would expect possibly indeterministic stuff like that when bad IO is beeing done)
and: pykalman/datasets/ not copied when pykalman is installed with pip #66
So it seems this dataset does not come with pip-install automatically, which looks like your problem and should be easily handled (manually obtaining the data!). It is also compatible with hpaulj's wondering about that strange error (one could argue if there is no better error-handling possible).
The following gives an error
drop function testing();
CREATE FUNCTION testing()
RETURNS text
AS $$
import ctypes
try:
ctypes.windll.LoadLibrary("D:\\jcc.dll")
except:
import traceback
plpy.error(traceback.format_exc())
return ''
$$ LANGUAGE plpythonu;
select testing();
Error message:
ERROR: ('Traceback (most recent call last):\n File "<string>", line 5, in __plpython_procedure_testing_1517640\n File "D:\\Python26\\Lib\\ctypes\\__init__.py", line 431, in LoadLibrary\n return self._dlltype(name)\n File "D:\\Python26\\Lib\\ctypes\\__init__.py", line 353, in __init__\n self._handle = _dlopen(self._name, mode)\nWindowsError: [Error 126] The specified module could not be found\n',)
It works fine in a python interpretor.
Python 2.6.4 (r264:75708, Oct 26 2009, 08:23:19) [MSC v.1500 32 bit (Intel)] on
win32
Type "help", "copyright", "credits" or "license" for more information.
>>> import ctypes
>>> ctypes.windll.LoadLibrary("D:\\jcc.dll")
<WinDLL 'D:\jcc.dll', handle 410000 at 1d9cb10>
>>>
"The specified module could not be found" is one of those helpful error messages Windows emits that doesn't always mean what you think it means.
Windows will produce that message if the DLL you tried to load or any dll it depends on could not be found.
Since PostgreSQL runs in its own user account it has a different PATH to that which your interpreter runs in when you're testing. If jcc.dll depends on (say) c:\jccsupportfiles\aaa.dll and c:\jccsupportfiles is on your PATH but not the Pg server's PATH, that would explain your problem.
Try using Dependency Walker (depends.exe) to determine which DLLs your DLL requires and where they are. See if it's a PATH issue.
Rather than messing with the Pg server's PATH, consider just putting all the DLLs required by jcc.dll in the same directory as jcc.dll. IIRC Windows will always look in the same directory as the module it's loading first when it tries to load a module it depends on.
I have a series of experiments that were written for MATLAB, but recently we are trying to run them through Octave instead. I realize they are mostly compatible, but I have been running into a few problems, and none of the online FAQs or directions I have found have addressed these at all. It's complicated a bit because there are multiple .m files that interact; however, for now I am going to focus on the main program. Anyway, so when I try to run the file (MLP.m) through octave, I get the following errors in the Terminal window:
error: dir: expecting directory or filename to be a char array
error: called from:
error: /Applications/Octave.app/Contents/Resources/share/octave/3.2.3/m/miscellaneous/dir.m at line 128, column 5
error: /Applications/MATLAB_R2008a/toolbox/psychoacoustics/MLParameters.m at line 86, column 7
error: /Applications/MATLAB_R2008a/toolbox/psychoacoustics/MLP.m at line 9, column 3
The lines it is referencing are as follows:
1)
d = dir([cd myslash 'Experiments_MLP' myslash '*.m']);
2)
s = MLParameters;
What about these lines is incompatible with Octave? I can't find anything online that indicates that these won't work.
After that, the Terminal window gives me this batch of nonsense:
dyld: Library not loaded: /usr/X11/lib/libfreetype.6.dylib
Referenced from: /usr/X11R6/lib/libfontconfig.1.dylib
Reason: Incompatible library version: libfontconfig.1.dylib requires version 13.0.0 or later, but libfreetype.6.dylib provides version 10.0.0
dyld: Library not loaded: /usr/X11/lib/libfreetype.6.dylib
Referenced from: /usr/X11R6/lib/libfontconfig.1.dylib
Reason: Incompatible library version: libfontconfig.1.dylib requires version 13.0.0 or later, but libfreetype.6.dylib provides version 10.0.0
/Applications/Gnuplot.app/Contents/Resources/bin/gnuplot: line 71: 1077 Trace/BPT trap GNUTERM="${GNUTERM}" GNUPLOT_HOME="${GNUPLOT_HOME}" PATH="${PATH}" DYLD_LIBRARY_PATH="${DYLD_LIBRARY_PATH}" HOME="${HOME}" GNUHELP="${GNUHELP}" DYLD_FRAMEWORK_PATH="${DYLD_FRAMEWORK_PATH}" GNUPLOT_PS_DIR="${GNUPLOT_PS_DIR}" DISPLAY="${DISPLAY}" GNUPLOT_DRIVER_DIR="${GNUPLOT_DRIVER_DIR}" "${ROOT}/bin/gnuplot-4.2.6" "$#"
/Applications/Gnuplot.app/Contents/Resources/bin/gnuplot: line 71: 1083 Trace/BPT trap GNUTERM="${GNUTERM}" GNUPLOT_HOME="${GNUPLOT_HOME}" PATH="${PATH}" DYLD_LIBRARY_PATH="${DYLD_LIBRARY_PATH}" HOME="${HOME}" GNUHELP="${GNUHELP}" DYLD_FRAMEWORK_PATH="${DYLD_FRAMEWORK_PATH}" GNUPLOT_PS_DIR="${GNUPLOT_PS_DIR}" DISPLAY="${DISPLAY}" GNUPLOT_DRIVER_DIR="${GNUPLOT_DRIVER_DIR}" "${ROOT}/bin/gnuplot-4.2.6" "$#"
error: you must have gnuplot installed to display graphics; if you have gnuplot installed in a non-standard location, see the 'gnuplot_binary' function
I have GNUPlot installed, and I checked the gnuplot_binary function, which didn't give me any answers. GNUPlot is installed in my /Applications directory, along with Octave itself. Why shouldn't this work? The README file that came with GNUPlot didn't indicate a special directory for it to be installed in. What about the the dyld library not loaded errors? Is that related to the GNUPlot problem, or is it something else?
Anyway, thanks for your help
I know you already solved your problem, but if you have problems again here are some links with basic information about the differences between Matlab and Octave:
Porting programs from Matlab to Octave
Differences between Octave and MATLAB
Addressing your first error, it's easier to explain with an example:
dirName = '/some/path'; %# base directory
filesPath = fullfile(dirName, 'MLP', '*.m'); %# full path string
d = dir(filesPath); %# expand/enumerate files
for i=1:numel(d)
disp( d(i).name )
end
You also could have built the path using string concatenation yourself:
%# '/some/path/MLP/*.m'
filesPath = [dirName filesep 'MLP' filesep '*.m'];
The above should work for both MATLAB and Octave
I'm using the execnet package to allow communication between Python scripts interpreted by different Python interpreters.
The following code (test_execnet.py):
import execnet
for python_version in ('python', 'python3'):
try:
gw = execnet.makegateway("popen//python="+python_version)
ch = gw.remote_exec('channel.send(1/3)')
res = ch.receive()
print(python_version, ': ', res, sep ="")
except:
print('problems with ', python_version)
Runs perfectly in the command-line Terminal, showing the following output:
$ python3 test_execnet.py
python: 0
python3: 0.333333333333
However, if I try to run the same code from within the Eclipse IDE, I get the following error:
'import site' failed; use -v for traceback
Traceback (most recent call last):
File "<string>", line 1, in <module>
File "<string>", line 4, in <module>
File "<string>", line 2, in <module>
File "/Library/Frameworks/Python.framework/Versions/3.1/lib/python3.1/site-packages/execnet/gateway_base.py", line 8, in <module>
import sys, os, weakref
File "/Library/Frameworks/Python.framework/Versions/3.1/lib/python3.1/os.py", line 380, in <module>
from _abcoll import MutableMapping # Can't use collections (bootstrap)
File "/Library/Frameworks/Python.framework/Versions/3.1/lib/python3.1/_abcoll.py", line 54
class Hashable(metaclass=ABCMeta):
^
SyntaxError: invalid syntax
problems with python
problems with python3
NOTE:
Eclipse Version: 3.6.0
PyDev Interpreter configured for the project: python3
"Preferences/Interpreter - Python"'s Python Interpreters:
python (/usr/bin/python)
python3 (/Library/Frameworks/Python.Framework/Versions/3.1/Resources/Python.app/Contents/MacOS/Python
EDIT:
I write a code to show the os.environ like this:
for python_version in ('python', 'python3'):
try:
import os
for item in os.environ:
print(item, '= ', os.environ[item])
except:
print('problems with ', python_version)
I got the following outputs:
eclipse_output.txt
terminal_output.txt
A FileMerge comparison of the files can be found at eclipse_output.txt vs. terminal_output.pdf.
Any hints?
Thanks
seems like pydev does site-customizations and particularly modifies things for interactive/console usage (judging from a very quick skim of http://github.com/aptana/Pydev/blob/master/plugins/org.python.pydev/pysrc/pydev_sitecustomize/sitecustomize.py ). This is not useful or fitting for execnet-mediated processes.
You could try to "del os.environ['PYTHONPATH']" before you invoke execnet.makegateway, or, to be more careful, just delete the sitecustomize part of it.
hth,
holger
'import site' failed; use -v for traceback
I have seen that when python was unable to find its landmark. Which that indicates there is a PYTHONHOME problem.
Check out http://docs.python.org/using/cmdline.html#envvar-PYTHONHOME maybe eclipse is screwing your environment up.
Edit:
Looked at your env dumps, looks like eclipse is definitely messing with PYTHONPATH, which will cause your child python processes to not work correctly.
Basically what you have going on here is eclipse starts a python v2 instance with a PYTHONPATH pointing to the python v2 directories. Then you spawn a python v3 process which tries to load its landmark from the python v2 directories...
You need to find a way to have eclipse not mess with the PYTHONPATH. I am not sure what eclipse is trying to do by doing that, but it is certainly no friend when you want to spawn new python processes.