Get the value of some weights in a model trained by TensorFlow

I have trained a ConvNet model with TensorFlow, and I want to get a particular weight in layer. For example in torch7 I would simply access model.modules[2].weights. to get the weights of layer 2. How would I do the same thing in TensorFlow?

Solution 1:

In TensorFlow, trained weights are represented by tf.Variable objects. If you created a tf.Variable—e.g. called v—yourself, you can get its value as a NumPy array by calling sess.run(v) (where sess is a tf.Session).

If you do not currently have a pointer to the tf.Variable, you can get a list of the trainable variables in the current graph by calling tf.trainable_variables(). This function returns a list of all trainable tf.Variable objects in the current graph, and you can select the one that you want by matching the v.name property. For example:

# Desired variable is called "tower_2/filter:0".
var = [v for v in tf.trainable_variables() if v.name == "tower_2/filter:0"][0]

Solution 2:

2.0 Compatible Answer: If we build a Model using Keras Sequential API, we can get the Weights of the Model using the code mentioned below:

!pip install tensorflow==2.1

from tf.keras import Sequential

model = Sequential()

model.add(Conv2D(filters=conv1_fmaps, kernel_size=conv1_ksize,
                         strides=conv1_stride, padding=conv1_pad,
                         activation=tf.nn.relu, input_shape=(height, width, channels),
                    data_format='channels_last'))

model.add(MaxPool2D(pool_size = (2,2), strides= (2,2), padding="VALID"))

model.add(Dropout(0.25))

model.add(Flatten())

model.add(Dense(units = 32, activation = 'relu'))

model.add(Dense(units = 10, activation = 'softmax'))

model.summary()

print(model.trainable_variables) 

The Last Statement, print(model.trainable_variables), will return the Weights of the Model as shown below:

    [<tf.Variable 'conv2d/kernel:0' shape=(3, 3, 1, 32) dtype=float32>,
 <tf.Variable 'conv2d/bias:0' shape=(32,) dtype=float32>, <tf.Variable 
'dense/kernel:0' shape=(6272, 32) dtype=float32>, <tf.Variable 'dense/bias:0' 
shape=(32,) dtype=float32>, <tf.Variable 'dense_1/kernel:0' shape=(32, 10) 
dtype=float32>, <tf.Variable 'dense_1/bias:0' shape=(10,) dtype=float32>]

Solution 3:

So if you proceed this code step by step, you will first get a list of used/trainable variables. Then you could sort them in a list where you sort weight matrices/lists to variable names, just for example how you were possible to deal with that information.

vars = tf.trainable_variables()
print(vars) #some infos about variables...
vars_vals = sess.run(vars)
for var, val in zip(vars, vars_vals):
    print("var: {}, value: {}".format(var.name, val)) #...or sort it in a list....