Appendices¶
“The end of the book is the beginning of mastery.”
These appendices are designed to serve as your quick-reference toolkit and troubleshooting companion as you build real-world TensorFlow projects.
Appendix A: Tensor Shapes Cheat Sheet (TensorFlow Style)¶
| Shape Notation | Meaning | Example |
|---|---|---|
(batch_size,) |
1D vector (e.g., labels) | [32] |
(batch_size, features) |
2D input (e.g., tabular data) | [32, 10] |
(batch_size, height, width, channels) |
Image input (NHWC format) | [32, 28, 28, 1] |
(batch_size, time_steps, features) |
Sequences (e.g., RNN input) | [32, 100, 64] |
(vocab_size, embedding_dim) |
Word embeddings | [10000, 300] |
💡 Did you know? TensorFlow prefers NHWC format for convolutional layers: [batch, height, width, channels], while PyTorch uses NCHW by default.
Appendix B: TensorFlow vs PyTorch – Code Comparison¶
| Task | TensorFlow (Keras) | PyTorch |
|---|---|---|
| Dense Layer | tf.keras.layers.Dense(128) |
nn.Linear(in, out) |
| Activation | activation='relu' |
F.relu(x) |
| Loss Function | SparseCategoricalCrossentropy() |
nn.CrossEntropyLoss() |
| Optimizer | Adam(learning_rate=1e-3) |
optim.Adam(model.parameters()) |
| Training Loop | model.fit(x, y) |
for epoch in ...: optimizer.step() |
| Model Definition | Subclass tf.keras.Model |
Subclass nn.Module |
Appendix C: Debugging Tips for TensorFlow¶
| Issue | Cause | Fix |
|---|---|---|
Graph execution error |
Mixing eager + graph mode | Use @tf.function carefully |
ValueError: Shapes (None, 1) != (None, ) |
Shape mismatch | Check model output vs label shape |
| Model not learning | Wrong loss, optimizer, or learning rate | Try lr=1e-3 and experiment |
| GPU not being used | Device misconfiguration | Use tf.config.list_physical_devices('GPU') |
| Out-of-memory error (OOM) | Batch size too large or model too big | Reduce batch size or use mixed precision |
🗂 Appendix E: Full API Reference Crosswalk¶
“Knowing what’s possible is the first step to mastery.”
This appendix maps the most essential TensorFlow API classes and functions to their official documentation links, along with a quick summary of when and why you’d use each.
| Module | API | Summary | Docs |
|---|---|---|---|
| tf.Tensor | tf.constant, tf.Variable | Core tensor creation. Use constants for fixed data, variables for trainable parameters. tf.Tensor | tf.tensor |
| tf.keras | Sequential, Model, Layer | Core model-building classes. Sequential for stacks, subclass Model for custom. tf.keras | tf.keras |
| tf.keras.layers | Dense, Conv2D, LSTM, Flatten, Dropout | Building blocks of neural networks: fully connected, CNN, RNN, etc. tf.keras.layers | tf.keras.layers |
| tf.keras.losses | CategoricalCrossentropy, MSE, | Huber Standard loss functions for classification, regression, and robust fitting. tf.keras.losses | tf.keras.losses |
| tf.keras.optimizers | Adam, SGD, RMSprop | Optimizers for gradient descent-based training. tf.keras.optimizers | tf.keras.optimizers |
| tf.data | Dataset.from_tensor_slices, .batch(), .shuffle() | Efficient, scalable data pipelines for training. tf.data | tf.data |
| tf.image | resize, random_flip, per_image_standardization | Image preprocessing utilities (augmentation, normalization). tf.image | tf.image |
| tf.function | @tf.function | Converts Python code into efficient graph-based TensorFlow execution. tf.function | tf.function |
| tf.GradientTape | tf.GradientTape() | Enables automatic differentiation for custom training loops. tf.GradientTape | tf.GradientTape |
| tf.lite | TFLiteConverter, Interpreter | Tools to convert and run models on mobile/embedded devices. tf.lite | tf.lite |
| tensorflow_recommenders | tfrs.Model, tfrs.tasks.Retrieval | TensorFlow Recommenders for personalized ranking and retrieval. TensorFlow Recommenders | TensorFlow Recommmenders |
| transformers (HF) | TFAutoModel, AutoTokenizer, pipeline() | Hugging Face Transformers (TF version) for NLP & vision models. Hugging Face Transformers | Hugging Face Transformers |
| tfx | ExampleGen, Trainer, Pusher, Evaluator | TensorFlow Extended pipeline components for production ML. TFX | TFX |
| tensorboard | tf.summary, TensorBoard callback | Logs scalars, images, and graphs during training. TensorBoard | TensorBoard |