Awesome
matio-ffi
A LuaJIT interface to MATIO
Installation
First, make sure MATIO is installed on your system. This package only requires the binary shared libraries (.so, .dylib, .dll). Please see your package management system to install MATIO. You can also download and compile matio from MATIO web page If you are manually compiling matio from scratch (instead of using your package manager), please disable hdf5 support, it seems to have a bug.
# OSX
brew install homebrew/science/libmatio
# Ubuntu <= 16.04
sudo apt-get install libmatio2
# Ubuntu 18.04
sudo apt-get install libmatio4
ln -s /usr/lib/x86_64-linux-gnu/libmatio.so.4 /usr/lib/x86_64-linux-gnu/libmatio.so
luarocks install matio
Usage
###Load a tensor from matlab array
local matio = require 'matio'
-- load a single array from file
tensor = matio.load('test.mat', 'var_a')
-- load multiple arrays from file
tensors = matio.load('test.mat',{'var1','var2'})
-- load all arrays from file
tensors = matio.load('test.mat')
###Load structs, cell arrays and strings
There is some basic support for more complex data structures defined by the MAT file format. Structs will be converted to lua tables, indexed by their field names. Cell arrays are converted to lua arrays (tables indexed by numerals).
local matio = require 'matio'
-- load a struct, which contains tensors, cell arrays and strings
my_complex_data = matio.load('test.mat')
If you want to read sequences of characters as lua strings instead of Torch CharTensor, enable the following flag:
matio.use_lua_strings = true
-- load data
data_with_strings = matio.load('test.mat')
###Save a tensor or a set of tensors to a .mat file
local matio = require 'matio'
-- save a single tensor to a .mat file
data = torch.rand(5,5)
matio.save('test1.mat',data)
-- save a set of tensors to a .mat file
data1 = torch.rand(5,5)
data2 = torch.rand(2,3):float()
matio.save('test2.mat',{t1=data1,t2=data2})
--save a mix of tensors, booleans, strings, and chars
data1 = torch.rand(2,3):float()
matio.save('test3.mat',{t1=data1,t2='hello',t3=9,t4=false})
--save a table of string keys as a struct
matio.save('test4.mat',{ myStruct = {t1=9,t2=false} })
By default, the matlab files are saved in MAT5 format, using ZLIB compression. To save without compression, change the following variable:
matio.compression = matio.ffi.COMPRESSION_NONE -- without compression
matio.compression = matio.ffi.COMPRESSION_ZLIB -- default compression
Calling MATIO C functions
All MATIO C functions are available in the matio.ffi.
namespace returned by require. The only difference is the naming, which is not prefixed
by Mat_
anymore.
For example, look at matio.load in init.lua