My Skill Lab
Machine LearningIntermediate

TensorFlow and Keras in Practice

Build, train and export production models with Keras.

Taught by Rohan Maharjan · ML Engineer

What you'll learn

  • Build models with the Keras Sequential and Functional APIs and train them with compile and fit
  • Construct efficient, reproducible tf.data pipelines with mapping, batching, caching and prefetching
  • Use callbacks for early stopping, checkpointing, learning-rate scheduling and TensorBoard logging
  • Train image classifiers with CNNs and transfer learning from Keras Applications
  • Build text classifiers with TextVectorization, embeddings and pretrained encoders from KerasHub
  • Export models to SavedModel, convert to LiteRT for mobile and serve them with TensorFlow Serving

About this course

Keras is one of the quickest routes from an idea to a trained, deployable neural network, and TensorFlow gives it a mature path to servers, phones and browsers. This practical course teaches the modern Keras 3 API on TensorFlow: Sequential and Functional models, tf.data input pipelines, callbacks such as early stopping and checkpoints, TensorBoard, convolutional and transfer-learning image models, text classification with TextVectorization and pretrained encoders, custom training steps, mixed precision, and finally exporting to SavedModel, LiteRT for mobile and TensorFlow Serving. Every lesson is hands-on with runnable code and real datasets. It is for developers and data scientists who know Python and the basics of machine learning and want to build production-ready deep learning models with TensorFlow and Keras. Prior deep learning experience helps but is not required; the first lesson rebuilds the essentials.

Course content

7 lessons
  1. 1Keras Essentials: Models, Compile and FitThe Sequential and Functional APIs, the compile-fit-evaluate cycle, and how Keras 3 sits on TensorFlow.Preview
  2. 2Input Pipelines With tf.dataBuild fast, reproducible data pipelines from arrays, files and generators using map, batch, cache and prefetch.
  3. 3Callbacks, Checkpoints and TensorBoardStop at the right epoch, save the best weights, schedule the learning rate and watch training live in TensorBoard.
  4. 4Image Models: CNNs and Transfer LearningTrain a convolutional network from scratch, then fine-tune EfficientNet from Keras Applications on your own photos.
  5. 5Text Models: Vectorisation, Embeddings and Pretrained EncodersClassify text with a TextVectorization layer inside the model, a bidirectional LSTM, and a pretrained BERT from KerasHub.
  6. 6Custom Training: Losses, Metrics and train_stepWrite custom losses and metrics, override train_step for full control, and speed up with mixed precision.
  7. 7Exporting and Serving ModelsSave in .keras and SavedModel formats, convert to LiteRT for mobile, serve with TensorFlow Serving and validate the artefact.