基于AI工程的Python

weights = [0.5, -1.2, 3.0]
alias = weights
alias[0] = 99.0
print(weights)

independent = list(weights)
independent[1] = 0.0
print(weights, independent)

grid = [[0] * 3] * 2
grid[0][0] = 1
print(grid)

grid = [[0] * 3 for _ in range(2)]
grid[0][0] = 1
print(grid)

[99.0, -1.2, 3.0]
[99.0, -1.2, 3.0] [99.0, 0.0, 3.0]
[[1, 0, 0], [1, 0, 0]]
[[1, 0, 0], [0, 0, 0]]

shape = (32, 784)
batch_size, n_features = shape
first, *middle, last = [1, 2, 3, 4, 5]
x, y = 1.0, 3.0
x, y = y, x

config = {"lr": 0.01, "epochs": 5}
config["seed"] = 7
tokens = ["the", "cat", "sat", "on", "the", "mat"]
vocab = set(tokens)

print(batch_size, n_features, first, middle, last, x, y)
print(config.get("momentum", 0.9), "cat" in vocab, len(vocab))
print(f"lr={config['lr']:.0e} loss={0.123456:.3f} n={60_000:,} {batch_size=}")

32 784 1 [2, 3, 4] 5 3.0 1.0
0.9 True 5
lr=1e-02 loss=0.123 n=60,000 batch_size=32

tokens = ["<s>", "the", "cat", "sat", "on", "the", "mat", "</s>"]
print(tokens[1:-1])
print(tokens[:3], tokens[-3:])
print(tokens[::2])
print(tokens[::-1][:2])
windows = [tokens[i:i + 3] for i in range(len(tokens) - 2)]
print(len(windows), windows[0])

['the', 'cat', 'sat', 'on', 'the', 'mat']
['<s>', 'the', 'cat'] ['the', 'mat', '</s>']
['<s>', 'cat', 'on', 'mat']
['</s>', 'mat']
6 ['<s>', 'the', 'cat']

scores = [0.91, 0.42, 0.77, 0.15, 0.66]
labels = [1 if s >= 0.5 else 0 for s in scores]
high = [s for s in scores if s > 0.6]
squares = {i: i * i for i in range(4)}
lengths = {len(word) for word in ["a", "bb", "cc", "ddd"]}
sum_sq = sum(s * s for s in scores)

matrix = [[1, 2, 3], [4, 5, 6]]
transposed = [list(column) for column in zip(*matrix)]
flat = [v for row in matrix for v in row]

print(labels, high, squares, lengths, round(sum_sq, 4))
print(transposed, flat)

[1, 0, 1, 0, 1] [0.91, 0.77, 0.66] {0: 0, 1: 1, 2: 4, 3: 9} {1, 2, 3} 2.0555
[[1, 4], [2, 5], [3, 6]] [1, 2, 3, 4, 5, 6]

def clip(value, low=0.0, high=1.0):
    return max(low, min(high, value))


def train(data, *, lr=0.01, epochs=3):
    return f"{len(data)} samples, lr={lr}, epochs={epochs}"


def make_scaler(factor):
    def scale(values):
        return [v * factor for v in values]
    return scale


print(clip(1.7), clip(-3.0, low=-1.0), clip(0.4, high=0.3))
print(train([1, 2, 3], lr=0.1))
to_percent = make_scaler(100)
print(to_percent([0.25, 0.5]))

1.0 -1.0 0.3
3 samples, lr=0.1, epochs=3
[25.0, 50.0]

def add_bad(item, bucket=[]):
    bucket.append(item)
    return bucket


def add_good(item, bucket=None):
    if bucket is None:
        bucket = []
    bucket.append(item)
    return bucket


add_bad("a")
print(add_bad("b"))
add_good("a")
print(add_good("b"))

['a', 'b']
['b']

from dataclasses import dataclass, field


@dataclass(frozen=True)
class Example:
    features: tuple[float, ...]
    label: int


@dataclass
class TrainConfig:
    lr: float = 0.01
    epochs: int = 5
    hidden: list[int] = field(default_factory=lambda: [64, 32])

    def describe(self):
        return f"lr={self.lr} epochs={self.epochs} hidden={self.hidden}"


ex = Example((1.5, 2.0), 1)
print(ex)
print(ex == Example((1.5, 2.0), 1), len({ex, Example((1.5, 2.0), 1)}))
print(TrainConfig(epochs=10).describe())
try:
    ex.label = 0
except AttributeError as err:
    print(type(err).__name__)

Example(features=(1.5, 2.0), label=1)
True 1
lr=0.01 epochs=10 hidden=[64, 32]
FrozenInstanceError

class SquaresDataset:
    def __init__(self, n):
        self.n = n

    def __len__(self):
        return self.n

    def __getitem__(self, index):
        if not 0 <= index < self.n:
            raise IndexError(index)
        return index, index * index


ds = SquaresDataset(4)
print(len(ds), ds[2])
print(list(ds))

4 (2, 4)
[(0, 0), (1, 1), (2, 4), (3, 9)]

numbers = [10, 20, 30]
it = iter(numbers)
print(next(it), next(it), next(it))
try:
    next(it)
except StopIteration:
    print("no more items")


def countdown(n):
    while n > 0:
        yield n
        n -= 1


gen = countdown(3)
print(type(gen).__name__)
print(list(gen))
print(list(gen))

10 20 30
no more items
generator
[3, 2, 1]
[]

import os
import tempfile


def read_rows(path):
    with open(path, encoding="utf-8") as fh:
        header = fh.readline().rstrip("\n").split(",")
        for line in fh:
            yield dict(zip(header, line.rstrip("\n").split(",")))


with tempfile.TemporaryDirectory() as tmp:
    path = os.path.join(tmp, "rows.csv")
    with open(path, "w", encoding="utf-8") as fh:
        fh.write("x,label\n")
        fh.writelines(f"{i * 0.5},{i % 2}\n" for i in range(5))
    rows = read_rows(path)
    print(next(rows))
    print(sum(1 for _ in rows))

{'x': '0.0', 'label': '0'}
4

import sys

as_list = [i * i for i in range(1_000_000)]
as_generator = (i * i for i in range(1_000_000))
print(sys.getsizeof(as_list), sys.getsizeof(as_generator))
print(sum(as_list) == sum(as_generator))

import time
from contextlib import contextmanager


@contextmanager
def timer(label):
    start = time.perf_counter()
    try:
        yield
    finally:
        print(f"{label}: {(time.perf_counter() - start) * 1000:.1f} ms")


def parse_label(text):
    try:
        return int(text)
    except ValueError:
        raise ValueError(f"label must be an integer, got {text!r}") from None


with timer("sum of squares"):
    total = sum(i * i for i in range(200_000))

print(parse_label("3"))
try:
    parse_label("cat")
except ValueError as err:
    print("rejected:", err)

sum of squares: 9.4 ms
3
rejected: label must be an integer, got 'cat'

from __future__ import annotations

from collections.abc import Iterable, Sequence


def mean(values: Sequence[float]) -> float:
    if not values:
        raise ValueError("mean() needs at least one value")
    return sum(values) / len(values)


def count_tokens(docs: Iterable[str]) -> int:
    return sum(len(doc.split()) for doc in docs)


def main() -> None:
    print(mean([1.0, 2.0, 4.0]))
    print(count_tokens(["the cat sat", "on the mat"]))
    print(mean.__annotations__)


if __name__ == "__main__":
    main()

2.3333333333333335
6
{'values': 'Sequence[float]', 'return': 'float'}

import csv
import json
import tempfile
from pathlib import Path

rows = [
    {"text": "great movie", "label": 1, "score": 0.93},
    {"text": "boring plot", "label": 0, "score": 0.12},
    {"text": "fine acting", "label": 1, "score": 0.58},
]

with tempfile.TemporaryDirectory() as tmp:
    json_path = Path(tmp) / "reviews.json"
    json_path.write_text(json.dumps(rows, indent=2), encoding="utf-8")
    from_json = json.loads(json_path.read_text(encoding="utf-8"))

    csv_path = Path(tmp) / "reviews.csv"
    with csv_path.open("w", newline="", encoding="utf-8") as fh:
        writer = csv.DictWriter(fh, fieldnames=["text", "label", "score"])
        writer.writeheader()
        writer.writerows(rows)
    with csv_path.open(newline="", encoding="utf-8") as fh:
        from_csv = list(csv.DictReader(fh))

print(from_json == rows)
print(from_csv[0])

by_score = sorted(rows, key=lambda r: r["score"], reverse=True)
print([r["text"] for r in by_score])
print(max(rows, key=lambda r: r["score"])["text"])
print([r["text"] for r in sorted(rows, key=lambda r: (-r["label"], r["text"]))])

True
{'text': 'great movie', 'label': '1', 'score': '0.93'}
['great movie', 'fine acting', 'boring plot']
great movie
['fine acting', 'great movie', 'boring plot']

from collections import Counter, defaultdict, deque
from itertools import chain, islice, product

labels = ["cat", "dog", "cat", "bird", "cat", "dog"]
counts = Counter(labels)
print(counts.most_common(2))

positions = defaultdict(list)
for i, label in enumerate(labels):
    positions[label].append(i)
print(dict(positions))

recent = deque(maxlen=3)
for loss in [2.0, 1.5, 1.2, 1.1, 0.9]:
    recent.append(loss)
print(list(recent), round(sum(recent) / len(recent), 3))

print(list(product([0.1, 0.01], [16, 32])))
print(list(chain([1, 2], [3], [4, 5])))
print(list(islice(range(100), 5, 10)))

[('cat', 3), ('dog', 2)]
{'cat': [0, 2, 4], 'dog': [1, 5], 'bird': [3]}
[1.2, 1.1, 0.9] 1.067
[(0.1, 16), (0.1, 32), (0.01, 16), (0.01, 32)]
[1, 2, 3, 4, 5]
[5, 6, 7, 8, 9]

import dis


def dot(a, b):
    total = 0.0
    for x, y in zip(a, b):
        total += x * y
    return total


dis.dis(dot)

import csv
import json
import tempfile
from dataclasses import dataclass
from pathlib import Path


@dataclass(frozen=True)
class Record:
    features: tuple[float, ...]
    label: int


def row_to_record(row, names, label_field="label"):
    if label_field not in row:
        raise ValueError(f"row has no {label_field!r} field: {row}")
    if set(row) != {*names, label_field}:
        raise ValueError(f"row fields {list(row)} do not match {[*names, label_field]}")
    return Record(tuple(float(row[name]) for name in names), int(row[label_field]))


def load_records(path, label_field="label"):
    path = Path(path)
    if path.suffix == ".json":
        rows = json.loads(path.read_text(encoding="utf-8"))
    elif path.suffix == ".csv":
        with path.open(newline="", encoding="utf-8") as fh:
            rows = list(csv.DictReader(fh))
    else:
        raise ValueError(f"unsupported file type: {path.suffix!r}")
    if not rows:
        return []
    names = [key for key in rows[0] if key != label_field]
    return [row_to_record(row, names, label_field) for row in rows]


with tempfile.TemporaryDirectory() as tmp:
    csv_path = Path(tmp) / "data.csv"
    csv_path.write_text("x1,x2,label\n1.5,60,1\n1.8,82.5,0\n", encoding="utf-8")
    json_path = Path(tmp) / "data.json"
    json_path.write_text('[{"x1": 1.5, "x2": 60, "label": 1}]', encoding="utf-8")
    print(load_records(csv_path))
    print(load_records(json_path))

[Record(features=(1.5, 60.0), label=1), Record(features=(1.8, 82.5), label=0)]
[Record(features=(1.5, 60.0), label=1)]

import random


def train_test_split(items, test_fraction=0.2, seed=0):
    if not 0.0 < test_fraction < 1.0:
        raise ValueError(f"test_fraction must be between 0 and 1, got {test_fraction}")
    if len(items) < 2:
        raise ValueError(f"need at least 2 items to split, got {len(items)}")
    order = list(range(len(items)))
    random.Random(seed).shuffle(order)
    n_test = min(max(1, round(len(items) * test_fraction)), len(items) - 1)
    test = [items[i] for i in order[:n_test]]
    train = [items[i] for i in order[n_test:]]
    return train, test


items = list(range(10))
print(train_test_split(items, seed=1))
print(train_test_split(items, seed=1) == train_test_split(items, seed=1))
print(train_test_split(items, seed=2)[1])
print(items)

([9, 7, 5, 3, 0, 4, 1, 2], [6, 8])
True
[5, 9]
[0, 1, 2, 3, 4, 5, 6, 7, 8, 9]

from itertools import islice


def batches(items, batch_size, drop_last=False):
    if batch_size < 1:
        raise ValueError(f"batch_size must be at least 1, got {batch_size}")
    return _batch_iter(iter(items), batch_size, drop_last)


def _batch_iter(iterator, batch_size, drop_last):
    while True:
        batch = list(islice(iterator, batch_size))
        if not batch or (drop_last and len(batch) < batch_size):
            return
        yield batch


print(list(batches(range(10), 4)))
print(list(batches(range(10), 4, drop_last=True)))
squares = (i * i for i in range(7))
print([len(b) for b in batches(squares, 3)])

for epoch in range(2):
    sizes = [len(b) for b in batches(range(10), 4)]
    print(f"epoch {epoch}: {sizes}")

[[0, 1, 2, 3], [4, 5, 6, 7], [8, 9]]
[[0, 1, 2, 3], [4, 5, 6, 7]]
[3, 3, 1]
epoch 0: [4, 4, 2]
epoch 1: [4, 4, 2]

import math


def dot(a, b):
    return sum(x * y for x, y in zip(a, b, strict=True))


def mean(values):
    if not values:
        raise ValueError("mean() needs at least one value")
    return sum(values) / len(values)


def std(values):
    mu = mean(values)
    return math.sqrt(sum((v - mu) ** 2 for v in values) / len(values))


def make_standardizer(rows):
    columns = list(zip(*rows))
    means = [mean(c) for c in columns]
    scales = [s if s > 0 else 1.0 for s in (std(c) for c in columns)]

    def standardize(row):
        return [(v - m) / s for v, m, s in zip(row, means, scales, strict=True)]

    return standardize


print(dot([1.0, 2.0, 3.0], [4.0, 5.0, 6.0]))
values = [2.0, 4.0, 4.0, 4.0, 5.0, 5.0, 7.0, 9.0]
print(mean(values), std(values))

train_rows = [(1.0, 100.0), (2.0, 300.0), (3.0, 500.0)]
standardize = make_standardizer(train_rows)
print([[round(v, 3) for v in standardize(r)] for r in train_rows])
print([round(v, 3) for v in standardize((4.0, 700.0))])

32.0
5.0 2.0
[[-1.225, -1.225], [0.0, 0.0], [1.225, 1.225]]
[2.449, 2.449]

python3 phases/00-setup-and-tooling/13-python-for-ai-engineering/code/main.py
npx tsx phases/00-setup-and-tooling/13-python-for-ai-engineering/code/main.ts

Python for AI engineering: a standard-library toolkit
data     400 records, CSV and JSON load back unchanged: True
split    320 train, 80 test, records in both: 0
batches  size 48: [48, 48, 48, 48, 48, 48, 32]
scale    train feature means after standardizing: [0.0, 0.0]
model    nearest-centroid test accuracy: raw 0.86, standardized 0.89
speed    dot product of 200,000 floats, best of 5:
           index loop         6.60 ms
           zip generator      5.93 ms
           map + mul          2.89 ms
memory   peak bytes to sum 100,000 squares:
           list              4,000,616
           generator               472

import operator
import random
import time
import tracemalloc


def dot_index_loop(a, b):
    total = 0.0
    for i in range(len(a)):
        total += a[i] * b[i]
    return total


def dot_zip(a, b):
    return sum(x * y for x, y in zip(a, b, strict=True))


def dot_map(a, b):
    return sum(map(operator.mul, a, b))


def best_time(fn, *args, repeat=5):
    times = []
    for _ in range(repeat):
        start = time.perf_counter()
        fn(*args)
        times.append(time.perf_counter() - start)
    return min(times)


def peak_bytes(fn):
    tracemalloc.start()
    try:
        fn()
        return tracemalloc.get_traced_memory()[1]
    finally:
        tracemalloc.stop()


rng = random.Random(0)
a = [rng.random() for _ in range(1_000_000)]
b = [rng.random() for _ in range(1_000_000)]
for fn in (dot_index_loop, dot_zip, dot_map):
    print(f"{fn.__name__:15} {best_time(fn, a, b) * 1000:7.2f} ms")

print(peak_bytes(lambda: sum([i * i for i in range(100_000)])))
print(peak_bytes(lambda: sum(i * i for i in range(100_000))))

import statistics
import sys

values = [2.0, 4.0, 4.0, 4.0, 5.0, 5.0, 7.0, 9.0]
print(statistics.fmean(values), statistics.pstdev(values))

if sys.version_info >= (3, 12):
    import math
    from itertools import batched

    print(math.sumprod([1.0, 2.0, 3.0], [4.0, 5.0, 6.0]))
    print(list(batched(range(10), 4)))

5.0 2.0
32.0
[(0, 1, 2, 3), (4, 5, 6, 7), (8, 9)]

import time

import numpy as np

n = 1_000_000
rng = np.random.default_rng(0)
a = rng.random(n)
b = rng.random(n)
a_list, b_list = a.tolist(), b.tolist()

start = time.perf_counter()
slow = sum(x * y for x, y in zip(a_list, b_list))
python_ms = (time.perf_counter() - start) * 1000

start = time.perf_counter()
fast = float(a @ b)
numpy_ms = (time.perf_counter() - start) * 1000

print(f"pure Python {python_ms:8.2f} ms")
print(f"NumPy       {numpy_ms:8.2f} ms")
print(f"difference  {abs(slow - fast):.1e}")
print(a.dtype, a.itemsize, a.nbytes)

import numpy as np

X = np.array([[1.0, 100.0], [2.0, 300.0], [3.0, 500.0]])
mu = X.mean(axis=0)
sigma = X.std(axis=0)
Z = (X - mu) / sigma
print(X.shape, mu.shape, Z.shape)
print(Z.round(3))

norms = np.sqrt((X ** 2).sum(axis=1, keepdims=True))
print(norms.shape, (X / norms).shape)

(3, 2) (2,) (3, 2)
[[-1.225 -1.225]
 [ 0.     0.   ]
 [ 1.225  1.225]]
(3, 1) (3, 2)