Commonly Used TensorFlow Funcitons

Neural Network Tutorials - Herong's Tutorial Examples

∟Commonly Used TensorFlow Funcitons

This section describes some commonly used TensorFlow functions, including add(), multiply(), matmul(), tf.reduce_sum(), tf.nn.softmax(), etc.

As a quick reference, here is a list of commonly used TensorFlow functions provided by different modules and classes

1. Functions provided by the "tensorflow" module itself:

import tensorflow as tf

tf.add(a, b) - Creates a tensor operation whose output tensor is calculated by taking the sum of elements from both input tensors at the same position. This is function is the same as expression (a + b), because the "+" operator is overloaded as tf.add().

tf.cast(a, t) - Creates a tensor operation whose output tensor is the input tensor, a, with data type changed to the given type, t.

tf.constant(cons) - Creates a tensor operation with an output tensor fixed to the given value.

tf.equal(a, b) - Creates a tensor operation with an output tensor of scalar Boolean true or false, if both input tensors are equal.

tf.exp(a) - Creates a tensor operation whose output tensor is calculated by applying Exp() on each element of the input tensor.

tf.matmul(a, b) - Creates a tensor operation whose output tensor is calculated by taking the dot product (matrix multiplication) of two input tensors.

tf.multiply(a, b) - Creates a tensor operation whose output tensor is calculated by taking the multiplication of elements from both input tensors at the same position. This is function is the same as expression (a * b), because the "*" operator is overloaded as tf.multiply().

tf.ones(s) - Creates a "constant" tensor operation whose output tensor is filled with 1 in the given dimension shape, s.

tf.one_hot(i, l, dtype=t) - Creates a "constant" tensor operation whose output tensor is filled with one-hot values of 1 on a list of given index locations, i, Each index is index location is expanded into a vector of length, l.

tf.reduce_mean(a, l) - Creates a tensor operation whose output tensor is the input tensor, a, with the given list of dimensions, l, reduced to its means (average values). If l is not specified, all dimensions are reduced and the output is a scalar.

tf.reduce_sum(a, l) - Creates a tensor operation whose output tensor is the input tensor, a, with the given list of dimensions, l, reduced to their sums. If l is not specified, all dimensions are reduced and the output is a scalar.

tf.sqrt(a) - Creates a tensor operation whose output tensor is calculated by taking the square root on each element of the input tensor.

tf.transpose(a) - Creates a tensor operation whose output tensor is the transposed version of the input tensor.

tf.variables_initializer(l) - Creates a special tensor operation whose job is to load initial values on the given list of "variable" tensors.

tf.zeros(s) - Creates a "constant" tensor operation whose output tensor is filled with 0 in the given dimension shape, s.

tf.nn.relu(a) - Creates a tensor operation whose output is calculated by applying the ReLU() activation function on each element in the input tensor

tf.nn.softmax(a, l) - Creates a tensor operation whose output is calculated by applying the Softmax() activation function on the input tensor, a, in the given list of dimensions, l. If l is not specified, the last dimension is used.

session = tf.Session() - Returns a tensor session object to be used to run (evaluate) any tensor expression represented as a tensor operation.

session.run(o, feed_dict=l) - Runs (evaluates) the given list of tensor operations, o, with the list of placeholders, l, to be loaded with specified values.

optimizer = tf.train.GradientDescentOptimizer(r) - Returns an optimizer object representing the Gradient Descent algorithm with the given learning rate, r.

optimizer.minimize(cost, var_list=l) - Creates a special tensor operation whose job is to update the given list of "variable" tensors using "optimizer" algorithm.