Reading H5 Groups and Attributes

Someone has ostensibly provided me with an h5 file (generated in python) that contains the weights of a neural network. These I need to open in matlab.

However, when I used the h5info and h5disp functions in MATLAB, the datasets part of the h5 file info is empty. However, there seems to be dataset information within the group category of the file, as you can see below.

How should I properly be accessing the weights with this h5 file?

Using data = hdf5read(filename,datasetname) seems to not work for no matter what I provide as the datasetname. I cannot provide the .h5 file since it is an unsupported file type with this forum.

Thank you very much

hinfo = 

      Filename: '/home/sem19h2/Documents/Karl/Semester_Thesis/Scripts/New_matlab_model/matlab_karl_v2/_weights.h5'
          Name: '/'
        Groups: [3x1 struct]
      Datasets: []
     Datatypes: []
         Links: []
    Attributes: [3x1 struct]

HDF5 _weights.h5 
Group '/' 
    Attributes:
        'layer_names':  'lstm_55 ', 'dense_29', 'lstm_56 '
        'backend':  'tensorflow'
        'keras_version':  '2.2.4'
    Group '/dense_29' 
        Attributes:
            'weight_names':  'dense_29/kernel:0', 'dense_29/bias:0  '
        Group '/dense_29/dense_29' 
            Dataset 'bias:0' 
                Size:  70
                MaxSize:  70
                Datatype:   H5T_IEEE_F32LE (single)
                ChunkSize:  []
                Filters:  none
                FillValue:  0.000000
            Dataset 'kernel:0' 
                Size:  70x70
                MaxSize:  70x70
                Datatype:   H5T_IEEE_F32LE (single)
                ChunkSize:  []
                Filters:  none
                FillValue:  0.000000
    Group '/lstm_55' 
        Attributes:
            'weight_names':  'lstm_55/kernel:0          ', 'lstm_55/recurrent_kernel:0', 'lstm_55/bias:0            '
        Group '/lstm_55/lstm_55' 
            Dataset 'bias:0' 
                Size:  280
                MaxSize:  280
                Datatype:   H5T_IEEE_F32LE (single)
                ChunkSize:  []
                Filters:  none
                FillValue:  0.000000
            Dataset 'kernel:0' 
                Size:  280x10
                MaxSize:  280x10
                Datatype:   H5T_IEEE_F32LE (single)
                ChunkSize:  []
                Filters:  none
                FillValue:  0.000000
            Dataset 'recurrent_kernel:0' 
                Size:  280x70
                MaxSize:  280x70
                Datatype:   H5T_IEEE_F32LE (single)
                ChunkSize:  []
                Filters:  none
                FillValue:  0.000000
    Group '/lstm_56' 
        Attributes:
            'weight_names':  'lstm_56/kernel:0          ', 'lstm_56/recurrent_kernel:0', 'lstm_56/bias:0            '
        Group '/lstm_56/lstm_56' 
            Dataset 'bias:0' 
                Size:  16
                MaxSize:  16
                Datatype:   H5T_IEEE_F32LE (single)
                ChunkSize:  []
                Filters:  none
                FillValue:  0.000000
            Dataset 'kernel:0' 
                Size:  16x70
                MaxSize:  16x70
                Datatype:   H5T_IEEE_F32LE (single)
                ChunkSize:  []
                Filters:  none
                FillValue:  0.000000
            Dataset 'recurrent_kernel:0' 
                Size:  16x4
                MaxSize:  16x4
                Datatype:   H5T_IEEE_F32LE (single)
                ChunkSize:  []
                Filters:  none
                FillValue:  0.000000

[Said Murtaza Jallali]

hey i have the same problem, have you solved it yet???

[Said Murtaza Jallali]

hey i have the same problem, have you solved it yet???

[Said Murtaza Jallali]

hey i have the same problem, have you solved it yet???

[Said Murtaza Jallali]

hey i have the same problem, have you solved it yet???

[Jacek Dmochowski]

You need to provide the name of the group(s) to the second argument of h5read().

For example, for grabbing the weights of the "dense_29" layer, you could do:

dense_29_weights = h5read(h5_filename,'/dense_29/dense_29/kernel:0');

Hope this helps.

Answer 1

[Sivylla Paraskevopoulou]

You might find useful the following two examples on how to import weights from a separate file.

1. Import Keras Network and Weights

https://www.mathworks.com/help/deeplearning/ref/importkerasnetwork.html#mw_22020fac-57ab-4230-a08e-14279668fe34

2. Import Keras Network Architecture and Weights from Separate Fileshttps://www.mathworks.com/help/deeplearning/ref/importkeraslayers.html#mw_654788ee-756d-463e-b334-49e107e41283