💻 Why do we use Python?

1. Interpreted vs. Compiled Languages: Fast Feedback Loop

• Compiled Languages (C/C++/Java)
• Source → (traditional) machine code → execution
• Pros: native performance; optimizations possible
• Cons: long write → compile → run cycle; full rebuild required for every change

• Interpreted Languages (Python, Ruby, etc.)
• Source → bytecode (.pyc) → executed line-by-line in a virtual machine
• Pros: immediate execution after modification; convenient REPL experimentation
• Cons: slower than native; concurrency limitations (GIL)

• Python: A Compromise
• Bytecode compilation + interpretation avoids full parsing on each run
• Combines fast startup of compiled languages with short feedback loop of interpreters

1.1 How to Use the Python Interpreter (Supplement to Tutorial Chapter 2)

• Execution & Invocation Methods (REPL: “Read → Eval → Print → Loop”)
• Interactive Mode: Run python3.13 in the terminal to start the REPL environment immediately; execute code line by line with instant results
• Script Mode: Run python script.py to execute a file in batch; all input is parsed and executed before exit
• Inline Execution: Use python -c "command" to run short code snippets instantly
• Module Execution: Use python -m module_name to run a module as __main__, treating the module path like a script

2. Developer’s Perspective: Productivity vs. Performance Trade-off

• “Fast development” equals fast experimentation
• When prototyping ML models, a few lines of script can run data preprocessing → training → evaluation in one go
• Impossible to compare with writing, building, and debugging preprocessing logic in C/C++

• High-level data structures & rich standard/third-party libraries
• Built-in containers (lists, dicts, tuples) handle complex data manipulation in a single line
• Ecosystem (NumPy, Pandas, Scikit-learn, TensorFlow, PyTorch) lets you piece together code snippets for format conversion, numerical computation, and model training

• Disadvantages can be mitigated
• Slow parts can be offloaded to Cython/C extensions or addressed with PyPy JIT
• Need parallelism? Use a process pool or external distributed systems (Ray, Dask)

3. Best for Studying AI

1. Instant Gratification
• Inspect DataFrame output and visualize immediately in the REPL
• Launch ipdb or pdb for step-by-step debugging

2. Tutorials, Examples, Community
• Tens of thousands of Jupyter notebooks on official tutorials, Stack Overflow, and GitHub
• Directly reuse “I ran this code and got these results” examples or derive new experiments

3. Platform Independence
• Install and run the same way on Windows, macOS, and Linux
• Bytecode (.pyc) and PVM structure minimize hardware dependencies

4. Python Prioritizes Productivity Over Performance

• Fast development feedback loop, high-level data structures, vast libraries, portability, and community culture all come together

• Essential for quickly building everything from ML/DL model experiments to deployment scripts, data visualizations, and automated reports