Error when using tf.keras model

[HugoTex98]

Hello!

I have been trying to apply an analyzer on a tf.keras model, as exemplified below.

def ConnectomeCNN(input_shape, keep_pr=0.65, n_filter=32, n_dense1=64, n_classes=2):
bias_init = tf.constant_initializer(value=0.001)
input_1 = InputLayer(input_shape=input_shape, name='input')
conv1 = Conv2D(filters=n_filter , kernel_size=(1,input_shape[1]), strides=(1, 1),
                                  padding= "valid", activation="selu", 
                                  kernel_initializer="glorot_uniform",
                                  bias_initializer=bias_init, name="conv1", input_shape=input_shape)
dropout1 = Dropout(keep_pr, name="dropout1")
conv2 = Conv2D(filters=n_filter*2 , kernel_size=(input_shape[1],1), strides=(1, 1),
                                  padding= "valid", activation="selu", 
                                  kernel_initializer="glorot_uniform",
                                  bias_initializer=bias_init, name="conv2")
dropout2 = Dropout(keep_pr, name="dropout2")
reshape = Reshape((n_filter*2,), name="reshape")
dense1 = Dense(n_dense1, activation="selu", name="dense1", kernel_regularizer='l1_l2') #kernel_regularizer = regularizers.l1(0.0001))

if n_classes == 1:
    activation = "sigmoid"
else:
    activation = "softmax"
output = Dense(n_classes, activation=activation, name="output") 

model = tf.keras.models.Sequential([input_1,conv1, dropout1, conv2, dropout2,
                                reshape, dense1, output])
return model`

I am able to create the intended analyzer to the model in cause, but when I try to apply the analyzer to my data analysis = LRP_analyzer.analyze(data_FC) it gives me this error "AttributeError: 'Node' object has no attribute 'output_masks’”.

I am currently using Tensorflow 1.11.0 and Keras 2.2.4 versions.

Any idea about the possible origin of the error? Thanks in advance!

[adrhill]

Hi Hugo,

this is related to issue #227: iNNvestigate currently expects a keras Model instead of a tf.keras model. This will be fixed with the next release of iNNvestigate!

In the meantime, you can get your code running by using keras.models.Sequential instead of tf.keras.models.Sequential:

import keras
import numpy as np
import tensorflow as tf
from keras.layers import Conv2D, Dense, Dropout, InputLayer, Reshape

import innvestigate
import innvestigate.utils as iutils

def ConnectomeCNN(input_shape, keep_pr=0.65, n_filter=32, n_dense1=64, n_classes=2):
    bias_init = tf.constant_initializer(value=0.001)
    input_1 = InputLayer(input_shape=input_shape, name="input")
    conv1 = Conv2D(
        filters=n_filter,
        kernel_size=(1, input_shape[1]),
        strides=(1, 1),
        padding="valid",
        activation="selu",
        kernel_initializer="glorot_uniform",
        bias_initializer=bias_init,
        name="conv1",
        input_shape=input_shape,
    )
    dropout1 = Dropout(keep_pr, name="dropout1")
    conv2 = Conv2D(
        filters=n_filter * 2,
        kernel_size=(input_shape[1], 1),
        strides=(1, 1),
        padding="valid",
        activation="selu",
        kernel_initializer="glorot_uniform",
        bias_initializer=bias_init,
        name="conv2",
    )
    dropout2 = Dropout(keep_pr, name="dropout2")
    reshape = Reshape((n_filter * 2,), name="reshape")
    dense1 = Dense(
        n_dense1, activation="selu", name="dense1", kernel_regularizer="l1_l2"
    )  # kernel_regularizer = regularizers.l1(0.0001))

    if n_classes == 1:
        activation = "sigmoid"
    else:
        activation = "softmax"
    output = Dense(n_classes, activation=activation, name="output")

    model = keras.models.Sequential(
        [input_1, conv1, dropout1, conv2, dropout2, reshape, dense1, output]
    )
    return model

# Create dummy input data
input_shape = (10, 10, 3)
batch_size = 2
data_x = np.random.rand(batch_size, *input_shape)

# Create model without trailing softmax
model = ConnectomeCNN(input_shape)
model = iutils.keras.graph.model_wo_softmax(model)

# Analyze model
analyzer = innvestigate.create_analyzer("lrp.epsilon", model)
a = analyzer.analyze(data_x)

[HugoTex98]

Thank you so much @adrhill !!

All the best!!