How Artificial Intelligence actually works

pip install scikit-learn matplotlib

# =============================================================
# STEP 1 - Import the tools we need
# =============================================================
import random                                            # picks different digits each run
from sklearn.datasets import load_digits           # example images
from sklearn.model_selection import train_test_split  # splits the data
from sklearn.neural_network import MLPClassifier     # the neural network
import matplotlib.pyplot as plt                       # draws pictures

# =============================================================
# STEP 2 - Load the data
# Each image is an 8x8 grid, flattened into 64 brightness numbers.
# =============================================================
digits = load_digits()
X = digits.data      # the pixels  (the question)
y = digits.target    # the answers (0 to 9)

print("Number of images:", len(X))
print("Numbers per image:", len(X[0]))

# =============================================================
# STEP 3 - Split into a training set and a test set
# The model NEVER sees the test set while learning. That is how
# we find out whether it really learnt, or just memorised.
# =============================================================
X_train, X_test, y_train, y_test = train_test_split(
    X, y, test_size=0.2, random_state=42)

# =============================================================
# STEP 4 - Build the network, then train it
# hidden_layer_sizes=(32,) means one hidden layer of 32 neurons.
# max_iter=500 means run the training loop up to 500 times.
# =============================================================
model = MLPClassifier(hidden_layer_sizes=(32,), max_iter=500, random_state=42)
model.fit(X_train, y_train)

# =============================================================
# STEP 5 - Test it on images it has never seen before
# =============================================================
accuracy = model.score(X_test, y_test)
print(f"Accuracy on new images: {accuracy:.1%}")

# =============================================================
# STEP 6 - Look at ten predictions with your own eyes
# X_test[0] would show the SAME digit every single run, because
# random_state=42 above fixes the train/test split identically
# each time. random.sample here picks 10 different images from
# the test set on every run, so the digits on screen change even
# though the trained model itself does not.
# =============================================================
sample_indexes = random.sample(range(len(X_test)), 10)

fig, axes = plt.subplots(2, 5, figsize=(9, 4))
for ax, idx in zip(axes.flat, sample_indexes):
    guess = model.predict([X_test[idx]])[0]
    correct = "correct" if guess == y_test[idx] else "WRONG"
    ax.imshow(X_test[idx].reshape(8, 8), cmap="gray")
    ax.set_title(f"guess: {guess} ({correct})", fontsize=9)
    ax.axis("off")
plt.tight_layout()
plt.show()

# =============================================================
# STEP 7 - See the actual numbers the network is looking at
# This is the exact same data as the first picture above, just
# printed as raw numbers instead of drawn as an image --- this
# is what "784 numbers go into the network" from Section 02
# really looks like (64 numbers here, since these images are a
# smaller 8x8 practice set rather than full 28x28 MNIST).
# =============================================================
first_idx = sample_indexes[0]
print(f"\nRaw pixel values the network received for image {first_idx}:")
print(X_test[first_idx].reshape(8, 8).round(1))

from sklearn.feature_extraction.text import CountVectorizer
from sklearn.naive_bayes import MultinomialNB

# STEP 1 - the training data: examples plus their correct labels
messages = [
    "free money click now",        "win a free prize today",
    "claim your free cash",        "urgent claim your prize now",
    "free free free offer expires", "can we move the meeting",
    "see you at training tonight",  "did you finish the homework",
    "mum says dinner is ready",     "the meeting is at three"
]
labels = ["spam", "spam", "spam", "spam", "spam",
          "normal", "normal", "normal", "normal", "normal"]

# STEP 2 - turn words into numbers (the model cannot read text)
# Each message becomes a count of how often each word appears.
vectoriser = CountVectorizer()
X = vectoriser.fit_transform(messages)

# STEP 3 - train
model = MultinomialNB()
model.fit(X, labels)

# STEP 4 - try it on messages it has never seen
tests = ["free prize click here", "are you coming to training", "claim cash now"]
for t in tests:
    prediction = model.predict(vectoriser.transform([t]))[0]
    print(t, "-->", prediction)