Computational Graphs and Static vs Dynamic Execution

Computational graphs are the fundamental abstraction that deep learning frameworks use to represent mathematical operations and their dependencies. The distinction between static and dynamic execution modes addresses a critical challenge in machine learning: how to optimize computation while maintaining flexibility for development.

In TensorFlow 1.x, static graphs were the default. Operations were defined in a graph structure first, then executed separately, enabling compiler-level optimizations and distributed execution across devices. However, this approach created friction for debugging and iteration because developers could not inspect intermediate values during development.

TensorFlow 2.x introduced eager execution as the default, where operations execute immediately, providing dynamic behavior similar to NumPy. This shift fundamentally changed how developers interact with the framework, trading some optimization potential for immediate feedback and Pythonic semantics.

Understanding both execution modes is critical because modern production systems often use graph mode for performance while development still leverages eager execution. High-performance applications frequently require explicit conversion between the two modes. The ability to switch between them—and to use @tf.function to explicitly compile eager code into graphs—gives practitioners powerful tools for balancing development speed with inference and training performance.

Analogy🏏Cricket

🏏 Think of it like cricket: Imagine Virat Kohli batting in a Test match innings—each delivery he faces builds on the context of all previous deliveries in that innings. The bowler's strategy evolves based on what happened in earlier overs; Kohli's mental state and approach shift based on the match situation, the bowler's previous deliveries, and the scoring rate. His decision to play an aggressive shot or defend depends entirely on this accumulated context—information from the past 50 deliveries that his mind actively maintains. Now map this to an RNN: each timestep is like one delivery Kohli faces, the input is the ball characteristics, the hidden state is Kohli's accumulated mental model of the bowler and match situation, and the output is his batting decision for that delivery. The recurrent connection is Kohli carrying forward his understanding from delivery 1 through delivery 2, 3, 4... all the way to delivery 50—he never resets this knowledge. However, vanilla RNNs suffer a critical problem: like a batsman whose memory of early overs fades by the 50th over (vanishing gradient), the network forgets distant context. LSTMs fix this like Kohli maintaining a written scorecard—explicit gates (input gate, forget gate, output gate) are like decision checkpoints where he consciously updates what he remembers (forget gate), what new information to integrate (input gate), and what to use for his next shot (output gate). This gating mechanism prevents information decay, allowing Kohli to maintain crucial context from delivery 1 even when deciding his shot on delivery 50. Understanding RNNs and LSTMs reveals why sequential problems fundamentally require mechanisms to preserve and selectively use historical information—just as Kohli's effectiveness depends on never losing track of the match narrative.