#_ the Tensorflow Foundational List
1. Basics:
● import tensorflow as tf: Import TensorFlow module.
● tf.__version__: Check TensorFlow version.
● tf.constant(value): Creates a constant tensor.
● tf.Variable(value): Creates a variable tensor.
● tf.zeros(shape): Create a tensor with all zeros.
● tf.ones(shape): Create a tensor with all ones.
2. Operations:
● tf.add(x, y): Add two tensors.
● tf.subtract(x, y): Subtract two tensors.
● tf.multiply(x, y): Multiply two tensors.
● tf.divide(x, y): Divide two tensors.
● tf.pow(x, y): X raised to power Y.
● tf.exp(x): Exponential of X.
● tf.sqrt(x): Square root of X.
3. Shapes & Reshaping:
● tf.shape(tensor): Get shape of a tensor.
● tf.reshape(tensor, shape): Reshape a tensor.
● tf.transpose(tensor): Transpose matrix.
● tf.squeeze(tensor): Removes dimensions of size 1.
● tf.expand_dims(tensor, axis): Inserts a new axis.
4. Data & I/O:
● tf.data.Dataset.from_tensor_slices(data): Create a dataset from
tensors.
● dataset.batch(size): Batch the dataset.
● dataset.shuffle(buffer_size): Shuffle the dataset.
● tf.data.experimental.make_csv_dataset(): Load CSV data.
By: Waleed Mousa
5. Neural Network Components:
● tf.keras.layers.Dense(units, activation): Fully connected layer.
● tf.keras.layers.Conv2D(filters, kernel_size): 2D convolutional
layer.
● tf.keras.layers.MaxPooling2D(pool_size): Max pooling layer.
● tf.keras.layers.Dropout(rate): Dropout layer.
● tf.keras.layers.Flatten(): Flatten layer.
6. Model Management:
● model.compile(optimizer, loss, metrics): Compile the model.
● model.fit(data, labels, epochs): Train the model.
● model.evaluate(data, labels): Evaluate the model's performance.
● model.save(filepath): Save a model to a file.
● tf.keras.models.load_model(filepath): Load a model from a file.
7. Optimizers, Losses, and Metrics:
● tf.keras.optimizers.Adam(learning_rate): Adam optimizer.
● tf.keras.losses.SparseCategoricalCrossentropy(): Cross-entropy
loss.
● tf.keras.metrics.Accuracy(): Accuracy metric.
8. Advanced Layers and Operations:
● tf.keras.layers.BatchNormalization(): Batch normalization layer.
● tf.keras.layers.Embedding(input_dim, output_dim): Embedding layer.
● tf.image.resize(images, size): Resize images.
● tf.signal.stft(signals, frame_length, frame_step): Short-time
Fourier transform.
9. Gradient & Training:
● with tf.GradientTape() as tape: ...: Record operations for
automatic differentiation.
● tape.gradient(target, sources): Compute the gradient of target
w.r.t sources.
By: Waleed Mousa
● optimizer.apply_gradients(zip(gradients, variables)): Apply
gradients to variables.
10. Custom & Advanced Models:
● Define a custom layer: subclass tf.keras.layers.Layer.
● Define a custom model: subclass tf.keras.Model.
● @tf.function: Decorator to compile function into a TensorFlow
graph for faster execution.
11. Debugging & Inspection:
● tf.debugging.assert_shapes(): Assertion for tensor shapes.
● tf.print(tensor): Print the value of a tensor.
12. Distributed Training:
● tf.distribute.MirroredStrategy(): Synchronous training on multiple
GPUs.
13. Transfer Learning:
● tf.keras.applications.MobileNetV2(): Pre-trained MobileNetV2 model.
● base_model.trainable = False: Freeze the base model for transfer
learning.
14. Callbacks:
● tf.keras.callbacks.TensorBoard(log_dir): TensorBoard visualization
callback.
● tf.keras.callbacks.EarlyStopping(patience): Stop training when a
monitored metric has stopped improving.
15. Regularization:
● tf.keras.regularizers.l1(l=0.01): L1 regularization.
● tf.keras.regularizers.l2(l=0.01): L2 regularization.
By: Waleed Mousa
16. Functional API:
● input = tf.keras.Input(shape): Define model input.
● x = layer(input): Connect a layer to another.
17. Eager Execution:
● tf.executing_eagerly(): Check if running in eager mode.
18. Advanced Neural Network Components:
● tf.keras.layers.GRU(units): Gated Recurrent Unit.
● tf.keras.layers.LSTM(units): Long Short-Term Memory layer.
● tf.keras.layers.Bidirectional(layer): Bidirectional wrapper for
RNNs.
19. Mobile & Edge Deployment:
● TensorFlow Lite: Convert models for mobile and edge devices.
● tf.lite.TFLiteConverter.from_keras_model(model): Converter to
create TensorFlow Lite model.
20. Extended Backend Operations:
● tf.math.sin(x): Sin function.
● tf.linalg.inv(matrix): Inverse of a matrix.
● tf.reduce_mean(tensor): Compute the mean of tensor elements.
21. Performance & Optimization:
● tf.function: Convert functions to high-performance TensorFlow
graphs.
● tf.TensorArray(): TensorFlow equivalent of a list for use in
tf.function.
22. Generative Adversarial Networks (GAN):
● tf.keras.layers.LeakyReLU(): Leaky Rectified Linear Unit activation
function often used in GANs.
By: Waleed Mousa
23. Object Detection & Image Segmentation:
● TensorFlow Object Detection API: Set of utilities to perform
object detection.
● tf.image.draw_bounding_boxes(): Draw bounding boxes on batch of
images.
24. Sequence to Sequence & Attention Mechanisms:
● tf.keras.layers.Attention(): Dot-product attention mechanism
layer.
● tf.keras.layers.AdditiveAttention(): Additive attention mechanism
layer.
25. Transformers & NLP:
● tf.keras.layers.MultiHeadAttention(): Multi-head attention layer,
used in transformers.
● Tokenization & Sequence Padding: Convert text to numerical data.
26. Time Series & Forecasting:
● tf.keras.layers.Conv1D(filters, kernel_size): 1D convolutional
layer for sequences.
● tf.data.Dataset.window(size, shift, stride): Create a sliding
window over a dataset.
27. Reinforcement Learning:
● TensorFlow Agents: Library for reinforcement learning in
TensorFlow.
28. Model Interpretability:
● tf.keras.layers.Activation(activation): Applies an activation
function.
● Feature importance: Identifying important input features to the
model predictions.
By: Waleed Mousa
29. TensorFlow Hub:
● Pre-trained models and reusable parts for TensorFlow.
● hub.KerasLayer("URL"): Use a model from TensorFlow Hub in Keras.
30. TensorFlow Extended (TFX):
● End-to-end platform for deploying production machine learning
pipelines.
● Data validation, transformation, and model serving.
31. TensorFlow.js:
● tfjs.converters.save_keras_model(model, path): Save Keras model for
TensorFlow.js.
● Browser-based machine learning.
32. Miscellaneous:
● TensorFlow Datasets: Library to load and preprocess datasets.
● tf.autograph: Convert Python constructs into their TensorFlow
equivalents.
● tf.data.experimental.cardinality(dataset): Determine the
cardinality of a dataset.
● tf.saved_model.load(path): Load a SavedModel.
● tf.keras.utils.get_file(): Downloads a file from a URL.
● tf.keras.utils.plot_model(model): Plots a model as a graph.
● TensorFlow Serving: Library developed by Google to serve
TensorFlow models.
● TensorFlow Graphics: Computer graphics and 3D vision library.
● TensorFlow Quantum: Library for quantum machine learning.
By: Waleed Mousa
