Error in loading Keras Model

[ahmed.osa...@gmail.com]

Hi , 

I am using Keras pre-trained model when I am trying to load this model an error ('No model found in config file.') appears

here is the code which I use for loading the model

from keras.models import load_model

base_model=load_model('kandinsky_x.h5')

here is the error

Using TensorFlow backend.

Traceback (most recent call last):

File "/home/deeplearning/Desktop/PythonScripts/Load_Keras_model.py", line 2, in <module>

base_model=load_model('kandinsky_x.h5')

File "/usr/local/lib/python2.7/dist-packages/keras/models.py", line 238, in load_model

raise ValueError('No model found in config file.')

ValueError: No model found in config file.

models are found here: https://github.com/jayanthkoushik/neural-style-models

any help.

[Matias Valdenegro]

The most obvious thing to check is that these HDF5 files are model files. And obviously, they are not, so the error message is correct. These files contain the weights only, meaning they were saved using model.save_weights, not model.save, which would have saved the model architecture in the HDF5 file.

--
You received this message because you are subscribed to the Google Groups "Keras-users" group.
To unsubscribe from this group and stop receiving emails from it, send an email to keras-users+unsubscribe@googlegroups.com.
To view this discussion on the web, visit https://groups.google.com/d/msgid/keras-users/c8b451ac-8788-4f7a-b0e9-523a2452ec6a%40googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

[ahmed osama]

So, is there any way to change the weight model ( model.save_weights) in to the architecture model  (model.save) . Also i have tried to load it as weights  in this code but seems to be something missed 

import numpy as np
import h5py
import keras.backend as K
from keras.models import Sequential

def str_shape(x):
    return 'x'.join(map(str, x.shape))

def load_weights(model, filepath, lookup={}, ignore=[], transform=None, verbose=True):
    """Modified version of keras load_weights that loads as much as it can.
    Useful for transfer learning.
    read the weights of layers stored in file and copy them to a model layer.
    the name of each layer is used to match the file's layers with the model's.
    It is possible to have layers in the model that dont appear in the file..
    The loading stopps if a problem is encountered and the weights of the
    file layer that first caused the problem are returned.
    # Arguments
        model: Model
            target
        filepath: str
            source hdf5 file
        lookup: dict (optional)
            by default, the weights of each layer in the file are copied to the
            layer with the same name in the model. Using lookup you can replace
            the file name with a different model layer name, or to a list of
            model layer names, in which case the same weights will be copied
            to all layer models.
        ignore: list (optional)
            list of model layer names to ignore in
        transform: None (optional)
            This is an optional function that receives the list of weighs
            read from a layer in the file and the model layer object to which
            these weights should be loaded.
        verbose: bool
            high recommended to keep this true and to follow the print messages.
    # Returns
        weights of the file layer which first caused the load to abort or None
        on successful load.
    """
    if verbose:
        print 'Loading', filepath, 'to', model.name
    flattened_layers = model.layers
    with h5py.File(filepath, mode='r') as f:
        # new file format
        layer_names = [n.decode('utf8') for n in f.attrs['layer_names']]

        # we batch weight value assignments in a single backend call
        # which provides a speedup in TensorFlow.
        weight_value_tuples = []
        for name in layer_names:
            if verbose:
                print name,
            g = f[name]
            weight_names = [n.decode('utf8') for n in
                            g.attrs['weight_names']]
            if len(weight_names):
                weight_values = [g[weight_name] for weight_name in
                                 weight_names]
                if verbose:
                    print 'loading', ' '.join(
                            str_shape(w) for w in weight_values),
                target_names = lookup.get(name, name)
                if isinstance(target_names, basestring):
                    target_names = [target_names]
                # handle the case were lookup asks to send the same weight to multiple layers
                target_names = [target_name for target_name in target_names if
                                target_name == name or target_name not in layer_names]
                for target_name in target_names:
                    if verbose:
                        print target_name,
                    try:
                        layer = model.get_layer(name=target_name)
                    except:
                        layer = None
                    if layer:
                        # the same weight_values are copied to each of the target layers
                        symbolic_weights = layer.trainable_weights + layer.non_trainable_weights

                        if transform is not None:
                            transformed_weight_values = transform(weight_values, layer)
                            if transformed_weight_values is not None:
                                if verbose:
                                    print '(%d->%d)'%(len(weight_values),len(transformed_weight_values)),
                                weight_values = transformed_weight_values

                        problem = len(symbolic_weights) != len(weight_values)
                        if problem and verbose:
                            print '(bad #wgts)',
                        if not problem:
                            weight_value_tuples += zip(symbolic_weights, weight_values)
                    else:
                        problem = True
                    if problem:
                        if verbose:
                            if name in ignore or ignore == '*':
                                print '(skipping)',
                            else:
                                print 'ABORT'
                        if not (name in ignore or ignore == '*'):
                            K.batch_set_value(weight_value_tuples)
                            return [np.array(w) for w in weight_values]
                if verbose:
                    print
            else:
                if verbose:
                    print 'skipping this is empty file layer'
        K.batch_set_value(weight_value_tuples)


model = Sequential()

load_weights(model , 'kandinsky_x.h5')

and the output is 

Using TensorFlow backend.
Loading kandinsky_x.h5 to sequential_1
input_1 skipping this is empty file layer
reflectpadding2d_1 skipping this is empty file layer
convolution2d_1 loading 32x3x9x9 32 convolution2d_1 ABORT

On Wednesday, August 16, 2017 at 3:53:19 PM UTC+2, Matias Valdenegro wrote:

The most obvious thing to check is that these HDF5 files are model files. And obviously, they are not, so the error message is correct. These files contain the weights only, meaning they were saved using model.save_weights, not model.save, which would have saved the model architecture in the HDF5 file.

On 16 August 2017 at 15:43, <ahmed.osa...@gmail.com> wrote:

Hi , 

I am using Keras pre-trained model when I am trying to load this model an error ('No model found in config file.') appears

here is the code which I use for loading the model

from keras.models import load_model

base_model=load_model('kandinsky_x.h5')

here is the error

Using TensorFlow backend.

Traceback (most recent call last):

File "/home/deeplearning/Desktop/PythonScripts/Load_Keras_model.py", line 2, in <module>

base_model=load_model('kandinsky_x.h5')

File "/usr/local/lib/python2.7/dist-packages/keras/models.py", line 238, in load_model

raise ValueError('No model found in config file.')

ValueError: No model found in config file.

models are found here: https://github.com/jayanthkoushik/neural-style-models

any help.

--
You received this message because you are subscribed to the Google Groups "Keras-users" group.

To unsubscribe from this group and stop receiving emails from it, send an email to keras-users...@googlegroups.com.