Introduction to TensorFlow

3. Introduction to TensorFlow

3.1 Dependent Views

Hosted on the GitHub platform and maintained by Google groups and contributors

Provide rich APIs related to deep learning, support Python and C/C++ interfaces

Provide a visual analysis tool Tensorboard for easy analysis and adjustment of the model

Supports various platforms such as Linux, Windows, Mac and even mobile devices

3.2 TensorFlow2.0 focuses on simplicity and ease of use

3.2.1 Workflow

• Load data using tf.data
• Model building and debugging
• Model training
• Pre-trained model call
• Deployment of the model

3.2.2 TensorFlow installation

pip install tensorflow==2.3.0 -i https://pypi.tuna.tsinghua.edu.cn/simple

3.3 Tensors

3.3.1 Tensors and related operations

• Tensor is a multidimensional array

  • 0-dimensional→scalar, one-dimensional tensor, two-dimensional, three-dimensional...

• Generate Tensor

  • tf.constant

• tensor to numpy

  • np.array()
  • Tensor.numpy()

• Common functions

  • Addition

    • tf.add

  • Element-wise multiplication

    • tf.muliply

  • Matrix Multiplication

    • tf.matmul

  • 

3.3.2 Variables

• A special kind of tensor whose shape is immutable, but whose parameters can be changed
• tf.Variable
• 

3.4 tf.keras

3.4.1 High-level API of TensorFlow2.0

• New styles and design patterns for TensorFlow code
• Improve the simplicity and reusability of TF code

3.4.2 Common Modules

3.4.3 Common Methods

• The main process of deep learning implementation

  • Data acquisition
  • Data processing
  • Model creation and training
  • Model testing and evaluation
  • Model predictions

• Import tf.keras

• Data entry

• Model building

• Training and Evaluation

• Callback function

• Model save and restore

3.5 Quick Start Model

3.5.1 Iris Classification Case

• Implementation using machine learning sklearn logistic regression

  • 

  • 

• Implementation using tf.Keras

  •