开源软件名称（OpenSource Name）：

开源软件地址(OpenSource Url)：

开源编程语言(OpenSource Language)：

开源软件介绍(OpenSource Introduction)：

DeepLocalize: Fault Localization for Deep Neural Networks

Our Tool

The simplest way to build a training model is to start with the [Sequential] model, our tool follows the Keras functional API, with some changes explained in the paper:

Here is the Sequential model:

model = Sequential()

Add all layers as easily as by using .add() function.

Dense layer takes three arguments: num_inputs(number of input unit), num_outputs(number of output unit), lr_rate(leaning rate), and name(name of layer).

The activation function is added as a new layer.

lr = 0.01 
model.add(Dense(num_inputs=100, num_outputs=64, lr_rate=lr, name='FC1'))
model.add(ReLu())
model.add(Dense(num_inputs=64, num_outputs=10, lr_rate=lr, name='FC2'))
model.add(softmax())

Once you finished building your model, you can use .compile() to start the learning process after determing the loss function, optimizer and the metrics:

model.compile(loss='categorical_crossentropy',
              optimizer='sgd',
              metrics=['accuracy'])

The core principle of our tool is to make the training model simple, while inserting instrumentation in the .fit() function to observe the model variables, then make the user to be fully in control when they need the variables.

model.fit(x_train, y_train, epochs=5, batch_size=32)

You can start training a new model using the following commands in Terminal (macOS/Linux) or cmd (Windows) as following:

python main.py

Our Callback Method

To use our callback, you need to add our callback as subclass in your keras.callbacks.py file.

The core principle of our callback to get a view on internal states and statistics of the model during training.

Then you can pass our callback DeepLocalize() to the .fit() method of a model as following:

callback = keras.callbacks.DeepLocalize(inputs, outputs, layer_number, batch_size, startTime)
model = keras.models.Sequential()
model.add(keras.layers.Dense(64))
model.add(keras.layers.Activation(activations.relu))
model.compile(keras.optimizers.SGD(), loss='mse')
model.fit(np.arange(100).reshape(5, 20), np.zeros(5), epochs=10, batch_size=1, 
...                     callbacks=[callback], verbose=0)

Prerequisites

Version numbers below are of confirmed working releases for this project.

python 3.6.5
Keras  2.2.0
Keras-Applications  1.0.2
Keras-Preprocessing 1.0.1  
numpy 1.19.2
pandas 1.1.5
scikit-learn 0.21.2
scipy 1.6.0
tensorflow 1.14.0

BibTeX Reference

If you find this paper useful in your research, please consider citing:

@inproceedings{wardat21DeepLocalize,
Author = {Mohammad Wardat and Wei Le and Hridesh Rajan},
Title = {DeepLocalize: Fault Localization for Deep Neural Networks},
Booktitle  = {ICSE'21: The 43nd International Conference on Software Engineering},
Year = {2021},
entrysubtype = {conference}
}