TensorFlow Cheat Sheet

Updated 2026-04-27

Next Topic: Time Series Forecasting Cheat Sheet

🧠Study flashcards on this topic145 cards · spaced repetition→

TensorFlow is Google's open-source machine learning framework designed for building, training, and deploying neural networks at scale. It combines low-level tensor operations with high-level Keras APIs, making it suitable for both research and production. TensorFlow runs on CPUs, GPUs, and TPUs, supports distributed training, and enables deployment to mobile, web, edge devices, and servers via TFLite, TF.js, and TF Serving. The key mental model: TensorFlow executes eager operations by default while @tf.function traces them into optimized computational graphs, with automatic differentiation via GradientTape handling backpropagation for training.

What This Cheat Sheet Covers

This topic spans 32 focused tables and 221 indexed concepts. Below is a complete table-by-table outline of this topic, spanning foundational concepts through advanced details.

Table 1: Model Building APIsTable 2: Core Layers - Dense and ConvolutionalTable 3: Reshaping and Utility LayersTable 4: Pooling LayersTable 5: Recurrent LayersTable 6: Normalization and Regularization LayersTable 7: Activation FunctionsTable 8: Tensor Creation OperationsTable 9: OptimizersTable 10: Loss FunctionsTable 11: MetricsTable 12: CallbacksTable 13: Learning Rate SchedulesTable 14: Data Loading and PreprocessingTable 15: Image Preprocessing LayersTable 16: Tensor Operations - Shape ManipulationTable 17: Mathematical OperationsTable 18: Variables and AssignmentTable 19: Gradient Computation and Custom TrainingTable 20: Model Training MethodsTable 21: Model PersistenceTable 22: Transfer Learning and Pre-trained ModelsTable 23: Custom Layers and ModelsTable 24: Embedding and Text LayersTable 25: Attention and Transformer LayersTable 26: Distributed Training StrategiesTable 27: Performance OptimizationTable 28: Model Deployment FormatsTable 29: Model Optimization ToolkitTable 30: Regularization TechniquesTable 31: Debugging and VisualizationTable 32: Advanced Data Pipeline Techniques

Quick IndexSubscribe to unlock

A jump-to index of every table row in this cheat sheet.

Mind MapSubscribe to unlock

An interactive map of every table and concept in this topic.

Table 1: Model Building APIs

Keras gives you three ways to define a model, and the right one depends on how unusual your architecture is. The Sequential API is the quickest path for a simple stack of layers, the Functional API handles branching graphs with multiple inputs, outputs, or shared layers, and subclassing trades convenience for total control when you need custom forward-pass logic. Most projects start Sequential and graduate as their needs grow.

API	Example	Description
Sequential API	`model = tf.keras.Sequential([` `tf.keras.layers.Dense(64, activation='relu'),` `tf.keras.layers.Dense(10, activation='softmax')` `])`	• Linear stack of layers where each has exactly one input and one output tensor • simplest approach for feedforward architectures.
Functional API	`inputs = tf.keras.Input(shape=(28, 28, 1))` `x = tf.keras.layers.Conv2D(32, 3)(inputs)` `outputs = tf.keras.layers.Dense(10)(x)` `model = tf.keras.Model(inputs, outputs)`	• Directed acyclic graph of layers • supports multiple inputs/outputs, shared layers, and non-linear topologies.

Table 1: Model Building APIs

API	Example	Description
Sequential API	`model = tf.keras.Sequential([` `tf.keras.layers.Dense(64, activation='relu'),` `tf.keras.layers.Dense(10, activation='softmax')` `])`	• Linear stack of layers where each has exactly one input and one output tensor • simplest approach for feedforward architectures.
Functional API	`inputs = tf.keras.Input(shape=(28, 28, 1))` `x = tf.keras.layers.Conv2D(32, 3)(inputs)` `outputs = tf.keras.layers.Dense(10)(x)` `model = tf.keras.Model(inputs, outputs)`	• Directed acyclic graph of layers • supports multiple inputs/outputs, shared layers, and non-linear topologies.