簡単なmatlab/Octaveスクリプトをpythonスクリプトに変換するツール
12866 ワード
http://stackoverflow.com/questions/9845292/a-tool-to-convert-matlab-code-to-python
これは既存のソースコードの書き換えに役立ちます.
しかし、プログラミングの観点からは、原生態のほうがいいはずだ.
OMPCは使いやすいですが、Python 2.5のサポートが一番いいです.
SMOPはずっと更新中(github)、これは現在最も良いはずです.Small Matlab Octave Pythonの頭文字の略語.
OctaveとMatlabの代わりにPythonを完全に使うことは期待できない.これも著者の初志ではない.ましてソフトウェアがこんなに小さいからといって、そんなに複雑な仕事を完成するのは難しい.
これはプラットフォームにまたがるコマンドラインで使用されるツールですが、Linuxでしかテストされていません.
SMOP
SMOP stands for Small Matlab/Octave to Python compiler. It is used to convert Matlab programs to Python.
SMOP is not a polished product, nor a replacement to Octave and Matlab. Taking into account its size (less than 3000 lines), this is not surprizing. There are no toolboxes. Small everyday functions (max, length, etc.) are recognized and supported, but that's all.
SMOP is written in Python, using PLY -- Python Lex/Yacc for lexical analysis and parsing, and numpy for runtime environment. SMOP is platform-independent, but is tested only on Linux. It is a command-line utility.
Example
It is possible to run an example without installing smop. Just unzip it somewhere, and cd there. In your current directory you will find a bunch of .py files and a file named fastsolver.m. It is taken from the winning submission to Matlab programming competition in 2004 (Moving Furniturehttp://www.mathworks.cn/matlabcentral/contest/contests/12/submissions/29989).
Now type python main.py fastsolver.m -o fastsolver.py. If you don't specify the output file with -o option, it is written to a.py. Each time a function is translated, its name is written.
The entire submission contains 2093 lines, and it is automatically translated to Python by smop. These are the good news. The bad news are that generating the code is not enough to run the program, so there are no performance numbers yet. While the submission itself --- the solver program --- does not use graphics, the envelope code that is responsible to run the submission, collect and display the results, does. So about 100 lines of the envelope must be rewritten by hand. Many standard functions are not yet implemented --- rand, find, and others. They are on the issues list. Some matlab constructs, especially creating arrays by out of bound assignment, are used in the submission, but not yet supported by smop. Meanwhile, these lines should be rewritten in the Matlab code.
becomes
The table below tries to summarize various features.
Implemented features
Lexical and syntactical analysis
Mostly complete, including some weird Matlab features
Name resolution
For each occurrence of a variable, find a set of its possible definitions
Inlining of small functions
Array subscripts translated from 1-based (Matlab and Fortran style) to 0-based (C and Python style)
Also, end subscript implemented
from:step:to translated to from:to:step
Upper bound is n+1
Unimplemented features
Structs
To be implemented as soon as cc possible.
Arrays silently become C=style (rows first).
In some cases it may break the code. Not detected.
Function handles and lambda expressions
Handles break the heuristic that tells between function calls and array references.
Graphics,
Never
Auto-expanding arrays
Unlike other languages, matlab allows out-of-bounds assignment. As MathWorks tries to phase out this feature, there is a lot of legacy code depending on it.
Sparse matrices
Have good chances of being implemented, especially taking into account that scipy have several implementations to choose from.
Full support for boolean indexing. Currently, some expressions don't work
For example, x(x>0.5) = 1 works, but y=x>0.5; x(y)=1 does not work.
Command syntax
Too complex to support
Type, rank and shape inference
Strings
up vote
22 down vote
accepted
There are several alternative tools for converting Matlab code to Python code (not tested yet):
Small Matlab to Python compiler: convert Matlab code to Python code, also developed here:SMOP@chiselapp
LiberMate: translate from Matlab to Python and SciPy
OMPC: Matlab to Python (a bit outdated)
Matlab to Python conversion: No download files available
Also, for those interested in an interface between the two languages and not conversion:
Python-Matlab wormholes: both directions of interaction supported
Python-Matlab bridge: use Matlab from within Python, offers matlab_magic for iPython, to execute normal matlab code from within ipython
PyMat: Control Matlab session from Python
MatPy: Python package for numerical linear algebra and plotting with a MatLab-like interface
Btw might be helpful to look here for other migration tips:
http://bci2000.org/downloads/BCPy2000/Migration.html
On a different note, though I'm not a
share
| edit
| flag
edited
Mar 20 at 8:26
Marco
104
1
12
answered
Jul 8 '13 at 20:45
johntex
602
5
12
add comment
これは既存のソースコードの書き換えに役立ちます.
しかし、プログラミングの観点からは、原生態のほうがいいはずだ.
OMPCは使いやすいですが、Python 2.5のサポートが一番いいです.
SMOPはずっと更新中(github)、これは現在最も良いはずです.Small Matlab Octave Pythonの頭文字の略語.
OctaveとMatlabの代わりにPythonを完全に使うことは期待できない.これも著者の初志ではない.ましてソフトウェアがこんなに小さいからといって、そんなに複雑な仕事を完成するのは難しい.
これはプラットフォームにまたがるコマンドラインで使用されるツールですが、Linuxでしかテストされていません.
SMOP
SMOP stands for Small Matlab/Octave to Python compiler. It is used to convert Matlab programs to Python.
SMOP is not a polished product, nor a replacement to Octave and Matlab. Taking into account its size (less than 3000 lines), this is not surprizing. There are no toolboxes. Small everyday functions (max, length, etc.) are recognized and supported, but that's all.
SMOP is written in Python, using PLY -- Python Lex/Yacc for lexical analysis and parsing, and numpy for runtime environment. SMOP is platform-independent, but is tested only on Linux. It is a command-line utility.
Example
It is possible to run an example without installing smop. Just unzip it somewhere, and cd there. In your current directory you will find a bunch of .py files and a file named fastsolver.m. It is taken from the winning submission to Matlab programming competition in 2004 (Moving Furniturehttp://www.mathworks.cn/matlabcentral/contest/contests/12/submissions/29989).
Now type python main.py fastsolver.m -o fastsolver.py. If you don't specify the output file with -o option, it is written to a.py. Each time a function is translated, its name is written.
lei@fuji:~/smop/smop$ python main.py fastsolver.m
fastsolver.m
solver
cbest
mainsolver
imoves
easysolver
localfiddler
findoverlaps
dijkstra
improve
TLL79
solverA
solver1
movefrompos
onemove
solver2
SearchPath
Faster10IntReps2
matrixsolver
outoftheway
ismember1
ismember2
setdiff
unique
sub2ind
randperm
perms
itTakesAThief
movefurniture
findshortestpath
dealWall1
lei@fuji:~/smop/smop$
The entire submission contains 2093 lines, and it is automatically translated to Python by smop. These are the good news. The bad news are that generating the code is not enough to run the program, so there are no performance numbers yet.
01 function moves=solver(A,B,w0)
02 [moves,optmove,optscore]=cbest(A,B,w0);
03 curscore=sum(w0(moves(:,1)));
04 lots=1;
05 if length(moves)-optmove<20||curscore/optscore<1.05
06 lots=2; return
07 else
08 lenw=length(w0);
09 [xx,nseq]=sort(rand(1,lenw));
10 A1=A;
11 B1=B;
12 w01=w0;
13 for i=1:lenw
14 A1(A==i)=nseq(i);
15 B1(B==i)=nseq(i);
16 w01(nseq(i))=w0(i);
17 end;
18 [moves2,optmove,optscore]=cbest(A1,B1,w01);
becomes
TBD
The table below tries to summarize various features.
Implemented features
Lexical and syntactical analysis
Mostly complete, including some weird Matlab features
Name resolution
For each occurrence of a variable, find a set of its possible definitions
Inlining of small functions
Array subscripts translated from 1-based (Matlab and Fortran style) to 0-based (C and Python style)
Also, end subscript implemented
from:step:to translated to from:to:step
Upper bound is n+1
Unimplemented features
Structs
To be implemented as soon as cc possible.
Arrays silently become C=style (rows first).
In some cases it may break the code. Not detected.
Function handles and lambda expressions
Handles break the heuristic that tells between function calls and array references.
Graphics,
Never
Auto-expanding arrays
Unlike other languages, matlab allows out-of-bounds assignment. As MathWorks tries to phase out this feature, there is a lot of legacy code depending on it.
Sparse matrices
Have good chances of being implemented, especially taking into account that scipy have several implementations to choose from.
Full support for boolean indexing. Currently, some expressions don't work
For example, x(x>0.5) = 1 works, but y=x>0.5; x(y)=1 does not work.
Command syntax
Too complex to support
Type, rank and shape inference
Strings
up vote
22 down vote
accepted
There are several alternative tools for converting Matlab code to Python code (not tested yet):
Small Matlab to Python compiler: convert Matlab code to Python code, also developed here:SMOP@chiselapp
LiberMate: translate from Matlab to Python and SciPy
OMPC: Matlab to Python (a bit outdated)
Matlab to Python conversion: No download files available
Also, for those interested in an interface between the two languages and not conversion:
pymatlab : communicate from Python by sending data to the MATLAB workspace, operating on them with scripts and pulling back the resulting data Python-Matlab wormholes: both directions of interaction supported
Python-Matlab bridge: use Matlab from within Python, offers matlab_magic for iPython, to execute normal matlab code from within ipython
PyMat: Control Matlab session from Python
pymat2 : continuation of the appearingly abandoned PyMat. mlabwrap , mlabwrap-purepy: make Matlab look like Python library (based on PyMat) oct2py : run GNU Octave commands from within Python pymex : Embeds the Python Interpreter in Matlab, also on File Exchange matpy : Access MATLAB in various ways: create variables, access .mat files, direct interface to MATLAB engine (requires MATLAB be installed). MatPy: Python package for numerical linear algebra and plotting with a MatLab-like interface
Btw might be helpful to look here for other migration tips:
http://bci2000.org/downloads/BCPy2000/Migration.html
On a different note, though I'm not a
fortran fan at all, for people who might find it useful there is: matlab2fortran share
| edit
| flag
edited
Mar 20 at 8:26
Marco
104
1
12
answered
Jul 8 '13 at 20:45
johntex
602
5
12
add comment