# Student coding labs Each program is independent and supplies its own imports, demo, input fixture or generator, and `--self-test` checks. Use Python 3.10 or newer. ## Setup ```sh python3 -m venv .venv source .venv/bin/activate python -m pip install -r requirements.txt ``` Windows PowerShell activation: `.venv\Scripts\Activate.ps1`. Sixteen programs need only the standard library. Five use PyTorch; PINN also uses NumPy and Matplotlib. The ML demos work on CPU and generate their data locally. ## Commands ### Python 101 — Basics for a New Bee ```sh python python_101.py python python_101.py --self-test ``` How do we turn CSV records into a useful command-line summary? ### The Science of Programming — Explained Simply ```sh python science_of_programming.py python science_of_programming.py --self-test ``` How can a contract determine the loop we should write? ### Weakest Preconditions — Explained Simply ```sh python weakest_preconditions.py python weakest_preconditions.py --self-test ``` Which starting states let a program finish with the required result? ### The Science of Programming — A Practical Python Playbook ```sh python science_of_programming_python.py python science_of_programming_python.py --self-test ``` How do we implement and check loops with precise boundary behavior? ### PyTorch 101 — From Tensors to Training ```sh python pytorch_101.py python pytorch_101.py --self-test ``` Can gradients learn the unknown slope and intercept of a line? ### Neural Networks and CNNs — From First Principles ```sh python cnn.py python cnn.py --self-test ``` Can shared local filters learn to distinguish vertical and horizontal bars? ### How LLMs Work — A Tech Talk ```sh python llm_tech_talk.py python llm_tech_talk.py --self-test ``` Can a small model learn next-character prediction while being unable to see future targets? ### Adaptive Neural Infrastructure with Memory Accumulation ```sh python adaptive_neural_infra.py python adaptive_neural_infra.py --self-test ``` Can we select experts, accumulate state, and keep memory scoped to a user? ### Physics-Informed Neural Networks ```sh python pinn_heat_equation.py --epochs 2500 python pinn_heat_equation.py --self-test ``` The PINN example creates synthetic heat-equation data if its CSV is absent, trains, saves a checkpoint, plots predictions/errors, and reports metrics. ### Automata Theory — DFA, NFA & Regular Expressions ```sh python automata_theory.py python automata_theory.py --self-test ``` Can a small state machine recognize a pattern one symbol at a time? ### Context-Free Grammars & Pushdown Automata ```sh python cfg_pda.py python cfg_pda.py --self-test ``` How does a stack or parsing table recognize nested structure? ### Turing Machines & Undecidability ```sh python turing_machines.py python turing_machines.py --self-test ``` How can a head and a transition table transform a binary number? ### P vs NP & NP-Completeness ```sh python p_vs_np.py python p_vs_np.py --self-test ``` How do we check a SAT witness and preserve satisfiability in a graph reduction? ### Lambda Calculus — Computation from Functions Alone ```sh python lambda_calculus.py python lambda_calculus.py --self-test ``` Can function application compute while preserving which binder a variable belongs to? ### Randomized Algorithms — Probability as a Tool ```sh python randomized_algorithms.py python randomized_algorithms.py --self-test ``` Which guarantees come from correctness, and which depend on random choices? ### Entropy & Shannon's Source Coding Theorem ```sh python entropy_source_coding.py python entropy_source_coding.py --self-test ``` Can a prefix code reduce the average bits needed for a skewed source? ### Channel Capacity & Error-Correcting Codes ```sh python channel_capacity_codes.py python channel_capacity_codes.py --self-test ``` How does structured redundancy correct a single flipped bit? ### KL Divergence, Cross-Entropy & Machine Learning ```sh python kl_cross_entropy.py python kl_cross_entropy.py --self-test ``` How do model probabilities turn into a stable loss and gradient? ### Kolmogorov Complexity — The Information in One Object ```sh python kolmogorov_complexity.py python kolmogorov_complexity.py --self-test ``` What can a short description show that a compressor's output length cannot? ### Compression in Practice — Arithmetic Coding, LZ & Beyond ```sh python compression_in_practice.py python compression_in_practice.py --self-test ``` Can we encode data compactly and decode it exactly? ### Monads in Python — With the Math ```sh python monads_python.py python monads_python.py --self-test ``` How can we compose successful steps while keeping missing values or errors explicit? ## Your own CSV ```sh python python_101.py --csv expenses.csv ``` Use the header `category,amount`, nonempty category names, and nonnegative decimal amounts with at most two decimal places. A sample `expenses.csv` is included. ## PINN outputs and prediction The default `output/` folder contains the trained checkpoint, two PNG plots, a loss-history CSV, and metrics JSON. Reusing this folder replaces generated outputs; choose a new `--output-dir` for each experiment. ```sh python pinn_heat_equation.py --predict 0.5 0.5 python pinn_heat_equation.py --data-weight 0 --output-dir output-physics-only ``` Synthetic CSV data must match `--alpha`. Use a new data path when changing diffusivity, as shown in the PINN presentation. ## Verification All 21 programs passed 6,220 checks and their demos ran from isolated temporary directories using Python 3.12 and PyTorch 2.14.1. Checks include exhaustive small-input enumeration, independent reference comparisons, codec round trips, shape contracts, numerical gradient checks, and training. Floating-point digits, compression sizes, and training results can vary across versions and hardware. Finite tests do not establish universal proofs. Theory examples explicitly retain their mathematical limits: a simulation budget does not decide halting, finite SAT enumeration does not settle P vs NP, and compressor sizes do not compute true Kolmogorov complexity.