TensorFlow Cheat Sheet

Here’s a cheat sheet for TensorFlow, a popular open-source machine learning library:

TensorFlow Basics

Import TensorFlow:

import tensorflow as tf

TensorFlow Version:

TensorFlow Version

Constants:

# Define a constant
x = tf.constant(5, name="x")

Variables:

# Define a variable
y = tf.Variable(10, name="y")

# Initialize variables
init = tf.compat.v1.global_variables_initializer()
with tf.compat.v1.Session() as sess:
    sess.run(init)

TensorFlow Operations

Addition:

z = x + y

Matrix Multiplication:

matrix_a = tf.constant([[1, 2], [3, 4]])
matrix_b = tf.constant([[5, 6], [7, 8]])
result = tf.matmul(matrix_a, matrix_b)

Activation Functions:

# Sigmoid
sigmoid_result = tf.nn.sigmoid(x)

# ReLU (Rectified Linear Unit)
relu_result = tf.nn.relu(x)

# Softmax
softmax_result = tf.nn.softmax(x)

TensorFlow Sessions

Session Creation:

with tf.compat.v1.Session() as sess:
    # Run operations
    result_value = sess.run(result)
    print(result_value)

TensorFlow Placeholder

Create a Placeholder:

# Placeholder for a vector of 3 elements
input_data = tf.compat.v1.placeholder(tf.float32, shape=(3,))

Feed Data into Placeholder:

input_values = [1.0, 2.0, 3.0]
with tf.compat.v1.Session() as sess:
    result = sess.run(z, feed_dict={x: 5, input_data: input_values})
    print(result)

TensorFlow Neural Network Example

# Define a simple neural network
input_features = tf.compat.v1.placeholder(tf.float32, shape=(None, 2), name="input_features")
weights = tf.Variable(tf.random.normal([2, 1]), name="weights")
bias = tf.Variable(tf.zeros([1]), name="bias")
output = tf.nn.sigmoid(tf.matmul(input_features, weights) + bias, name="output")

# Train the neural network
labels = tf.compat.v1.placeholder(tf.float32, shape=(None, 1), name="labels")
loss = tf.reduce_mean(tf.nn.sigmoid_cross_entropy_with_logits(logits=output, labels=labels), name="loss")
optimizer = tf.compat.v1.train.GradientDescentOptimizer(learning_rate=0.01)
train_op = optimizer.minimize(loss)

# Run a training session
with tf.compat.v1.Session() as sess:
    sess.run(tf.compat.v1.global_variables_initializer())
    for step in range(num_steps):
        _, current_loss = sess.run([train_op, loss], feed_dict={input_features: input_data, labels: labels_data})
        if step % display_step == 0:
            print(f"Step {step}, Loss={current_loss}")

This is a basic TensorFlow cheat sheet covering some fundamental operations. For more advanced topics and functionalities, refer to the official TensorFlow documentation.