errors when visualize_saliency using conv1d model about keras-vis

[raghakot]

Can you post the full gist of your code for me to test?

[zhangzhishan]

@raghakot After I updated to the latest version of the code, I still get error https://pastebin.com/tqUSsQjg.
My full code is

layer_name = 'activation_5'
layer_idx = [idx for idx, layer in enumerate(model.layers) if layer.name == layer_name][0]
# Here we are asking it to show attention such that prob of `pred_class` is maximized.
layer_output = model.layers[layer_idx].output
# print layer_output[:, 0]
heatmap = visualize_saliency(model, layer_idx, [pred_class], x)

My model struct is

[{'class_name': 'Conv1D', 'config': {'kernel_initializer': {'class_name': 'VarianceScaling', 'config': {'distribution': 'uniform', 'scale': 1.0, 'seed': None, 'mode': 'fan_avg'}}, 'name': u'convolution1d_1', 'kernel_constraint': None, 'bias_regularizer': None, 'bias_constraint': None, 'dtype': u'float32', 'activation': 'linear', 'trainable': True, 'filters': 32, 'padding': u'same', 'strides': (1,), 'dilation_rate': (1,), 'kernel_regularizer': None, 'bias_initializer': {'class_name': 'Zeros', 'config': {}}, 'batch_input_shape': (None, 1, 401), 'use_bias': True, 'activity_regularizer': None, 'kernel_size': (3,)}}, {'class_name': 'Activation', 'config': {'activation': 'relu', 'trainable': True, 'name': u'activation_1'}}, {'class_name': 'MaxPooling1D', 'config': {'padding': u'same', 'strides': (2,), 'trainable': True, 'name': u'maxpooling1d_1', 'pool_size': (2,)}}, {'class_name': 'Conv1D', 'config': {'kernel_initializer': {'class_name': 'VarianceScaling', 'config': {'distribution': 'uniform', 'scale': 1.0, 'seed': None, 'mode': 'fan_avg'}}, 'name': u'convolution1d_2', 'kernel_constraint': None, 'bias_regularizer': None, 'bias_constraint': None, 'dtype': 'float32', 'activation': 'linear', 'trainable': True, 'filters': 32, 'padding': u'same', 'strides': (1,), 'dilation_rate': (1,), 'kernel_regularizer': None, 'bias_initializer': {'class_name': 'Zeros', 'config': {}}, 'batch_input_shape': (None, None, None), 'use_bias': True, 'activity_regularizer': None, 'kernel_size': (3,)}}, {'class_name': 'Activation', 'config': {'activation': 'relu', 'trainable': True, 'name': u'activation_2'}}, {'class_name': 'MaxPooling1D', 'config': {'padding': u'same', 'strides': (2,), 'trainable': True, 'name': u'maxpooling1d_2', 'pool_size': (2,)}}, {'class_name': 'Conv1D', 'config': {'kernel_initializer': {'class_name': 'VarianceScaling', 'config': {'distribution': 'uniform', 'scale': 1.0, 'seed': None, 'mode': 'fan_avg'}}, 'name': u'convolution1d_3', 'kernel_constraint': None, 'bias_regularizer': None, 'bias_constraint': None, 'dtype': 'float32', 'activation': 'linear', 'trainable': True, 'filters': 64, 'padding': u'same', 'strides': (1,), 'dilation_rate': (1,), 'kernel_regularizer': None, 'bias_initializer': {'class_name': 'Zeros', 'config': {}}, 'batch_input_shape': (None, None, None), 'use_bias': True, 'activity_regularizer': None, 'kernel_size': (3,)}}, {'class_name': 'Activation', 'config': {'activation': 'relu', 'trainable': True, 'name': u'activation_3'}}, {'class_name': 'MaxPooling1D', 'config': {'padding': u'same', 'strides': (2,), 'trainable': True, 'name': u'maxpooling1d_3', 'pool_size': (2,)}}, {'class_name': 'Flatten', 'config': {'trainable': True, 'name': u'flatten_1'}}, {'class_name': 'Dense', 'config': {'kernel_initializer': {'class_name': 'VarianceScaling', 'config': {'distribution': 'uniform', 'scale': 1.0, 'seed': None, 'mode': 'fan_avg'}}, 'name': u'dense_1', 'kernel_constraint': None, 'bias_regularizer': None, 'bias_constraint': None, 'dtype': 'float32', 'activation': 'linear', 'trainable': True, 'kernel_regularizer': None, 'bias_initializer': {'class_name': 'Zeros', 'config': {}}, 'units': 64, 'batch_input_shape': (None, 64), 'use_bias': True, 'activity_regularizer': None}}, {'class_name': 'Activation', 'config': {'activation': 'relu', 'trainable': True, 'name': u'activation_4'}}, {'class_name': 'Dropout', 'config': {'rate': 0.5, 'trainable': True, 'name': u'dropout_1'}}, {'class_name': 'Dense', 'config': {'kernel_initializer': {'class_name': 'VarianceScaling', 'config': {'distribution': 'uniform', 'scale': 1.0, 'seed': None, 'mode': 'fan_avg'}}, 'name': u'dense_2', 'kernel_constraint': None, 'bias_regularizer': None, 'bias_constraint': None, 'dtype': 'float32', 'activation': 'linear', 'trainable': True, 'kernel_regularizer': None, 'bias_initializer': {'class_name': 'Zeros', 'config': {}}, 'units': 1, 'batch_input_shape': (None, 64), 'use_bias': True, 'activity_regularizer': None}}, {'class_name': 'Activation', 'config': {'activation': 'sigmoid', 'trainable': True, 'name': u'activation_5'}}]

The last layer is activation_5 using sigmoid activation function.

[raghakot]

This should now work. Can you try again?

[raghakot]

Closing. Feel free to open if this is still an issue.

[nassarofficial]

Hello, I'm having the same problem with a different model. Here is my model: https://gist.github.com/nassarofficial/ff2ca99114b2a4a9e07d336fe64dce71