import numpy as np
import pandas as pd
import matplotlib.pyplot as plt

rng = np.random.default_rng(3)

NAVY="#0a1230"; INK="#1a2138"; INK_SOFT="#4a5578"
CYAN="#0891b2"; PURPLE="#7c3aed"; AMBER="#d97706"
GREEN="#059669"; PINK="#db2777"; BLUE="#2563eb"
GRID="#e6e9f2"
plt.rcParams.update({
    "figure.facecolor":"white","axes.facecolor":"white","figure.dpi":110,"font.size":11,
    "axes.edgecolor":GRID,"axes.linewidth":1.2,"axes.grid":True,"grid.color":GRID,"axes.axisbelow":True,
    "axes.spines.top":False,"axes.spines.right":False,
    "axes.titlesize":15,"axes.titleweight":"bold","axes.titlecolor":INK,
    "axes.labelcolor":INK_SOFT,"axes.labelsize":11.5,"xtick.color":INK_SOFT,"ytick.color":INK_SOFT,"legend.frameon":False,
})
def titlecard(ax, title, subtitle=None):
    ax.set_title(title, loc="left", pad=18)
    if subtitle:
        ax.text(0, 1.02, subtitle, transform=ax.transAxes, fontsize=10.5, color=INK_SOFT, va="bottom")
print("✅ Environment ready.")

✅ Environment ready.

QUESTION = "Can we predict daily ice-cream sales from the day's temperature?"
TARGET   = "sales"        # what we want to predict (the label)
FEATURE  = "temperature"  # what we predict it from
METRIC   = "RMSE (typical error in number of ice creams) — lower is better"

print("QUESTION :", QUESTION)
print("PREDICT  :", TARGET, "   FROM:", FEATURE)
print("SUCCESS  :", METRIC)

QUESTION : Can we predict daily ice-cream sales from the day's temperature?
PREDICT  : sales    FROM: temperature
SUCCESS  : RMSE (typical error in number of ice creams) — lower is better

# 14 days of records (temperature in °C, ice creams sold). Real data is rarely clean:
raw = pd.DataFrame({
    "temperature": [18, 20, 21, 23, 25, 26, 28, 30, 31, 33, 25, 27, 22, 24],
    "sales":       [22, 28, 30, 35, 40, 44, 52, 60,  5, 70, 40, None, 26, 34],
})
print(f"Collected {len(raw)} rows.")
raw

Collected 14 rows.

data = raw.copy()
before = len(data)

# (a) DUPLICATES — drop exact repeat rows
data = data.drop_duplicates().reset_index(drop=True)
print(f"Removed {before - len(data)} duplicate row(s).")

# (b) MISSING VALUES — fill the gap with the average sales
fill_value = round(data["sales"].mean())
n_missing = int(data["sales"].isna().sum())
data["sales"] = data["sales"].fillna(fill_value)
print(f"Filled {n_missing} missing sales value(s) with the mean ({fill_value}).")

# (c) OUTLIERS — a hot day with almost no sales is a data-entry error; drop it
outliers = data[(data["temperature"] > 28) & (data["sales"] < 15)]
data = data.drop(outliers.index).reset_index(drop=True)
print(f"Removed {len(outliers)} outlier row(s).")
print(f"\nClean dataset: {len(data)} rows ready for analysis.")

Removed 1 duplicate row(s).
Filled 1 missing sales value(s) with the mean (37).
Removed 1 outlier row(s).

Clean dataset: 12 rows ready for analysis.

print(data[["temperature","sales"]].describe().round(1))

corr = data["temperature"].corr(data["sales"])
print(f"\nCorrelation(temperature, sales) = {corr:.2f}  -> strong, positive")

       temperature  sales
count         12.0   12.0
mean          24.8   39.8
std            4.3   14.4
min           18.0   22.0
25%           21.8   29.5
50%           24.5   36.0
75%           27.2   46.0
max           33.0   70.0

Correlation(temperature, sales) = 0.96  -> strong, positive

fig, ax = plt.subplots(figsize=(9,4.8))
ax.scatter(data["temperature"], data["sales"], color=GREEN, s=70, alpha=0.8, edgecolor="white", linewidth=1)
titlecard(ax, "Sales rise with Temperature", f"Exploratory scatter · correlation = {corr:.2f}")
ax.set_xlabel("Temperature (°C)"); ax.set_ylabel("Ice creams sold")
plt.tight_layout(); plt.show()

# Split into TRAIN (learn from) and TEST (judge on) — never evaluate on data you trained on
idx = rng.permutation(len(data))
test_idx  = idx[:4]
train_idx = idx[4:]
train, test = data.iloc[train_idx], data.iloc[test_idx]
print(f"Train on {len(train)} days, test on {len(test)} held-out days.")

# "Train" the model = fit the best straight line on the TRAIN data only
slope, intercept = np.polyfit(train["temperature"], train["sales"], 1)
print(f"Learned model:  sales = {intercept:.1f} + {slope:.1f} * temperature")

Train on 8 days, test on 4 held-out days.
Learned model:  sales = -34.3 + 3.0 * temperature

fig, ax = plt.subplots(figsize=(9,4.8))
ax.scatter(train["temperature"], train["sales"], color=CYAN, s=70, alpha=0.85, edgecolor="white", label="Training days")
xs = np.linspace(16, 35, 100)
ax.plot(xs, slope*xs+intercept, color=PURPLE, lw=2.6, label="Learned model")
titlecard(ax, "The Model Learns the Trend", "A straight line fit to the training data")
ax.set_xlabel("Temperature (°C)"); ax.set_ylabel("Ice creams sold"); ax.legend(loc="upper left")
plt.tight_layout(); plt.show()

# Evaluate on the TEST days the model never saw
pred_test = slope*test["temperature"] + intercept
rmse = np.sqrt(np.mean((test["sales"].values - pred_test.values)**2))
ss_res = np.sum((test["sales"].values - pred_test.values)**2)
ss_tot = np.sum((test["sales"].values - test["sales"].mean())**2)
r2 = 1 - ss_res/ss_tot
print(f"On unseen test days:  RMSE = {rmse:.1f} ice creams,   R² = {r2:.2f}")

# DEPLOY = wrap the model in a simple tool anyone can use
def predict_sales(temp_c):
    return max(0, round(slope*temp_c + intercept))

tomorrow = 29
print(f"\n📦 Forecast: tomorrow is {tomorrow}°C -> stock for about {predict_sales(tomorrow)} ice creams.")

On unseen test days:  RMSE = 4.1 ice creams,   R² = 0.90

📦 Forecast: tomorrow is 29°C -> stock for about 53 ice creams.

# Communicate: show predictions vs. reality on the test days
fig, ax = plt.subplots(figsize=(9,4.8))
ax.scatter(train["temperature"], train["sales"], color="#c7c9d9", s=45, alpha=0.7, edgecolor="white", label="train")
ax.scatter(test["temperature"], test["sales"], color=PINK, s=90, edgecolor="white", linewidth=1, zorder=5, label="actual (test)")
ax.plot(xs, slope*xs+intercept, color=PURPLE, lw=2.4, label="model")
ax.scatter([tomorrow],[predict_sales(tomorrow)], color=AMBER, s=200, marker="*", zorder=6, edgecolor="white", linewidth=1.2, label=f"forecast @ {tomorrow}°C")
titlecard(ax, "Does the Forecast Hold Up?", f"Tested on unseen days · RMSE = {rmse:.1f}")
ax.set_xlabel("Temperature (°C)"); ax.set_ylabel("Ice creams sold"); ax.legend(loc="upper left", fontsize=9)
plt.tight_layout(); plt.show()

Stage	What we do	Output
1 · Frame	Turn the business need into a data question + metric	a clear goal
2 · Collect	Gather the raw data	a raw dataset
3 · Clean	Fix duplicates, missing values, outliers	a tidy dataset
4 · Explore	Look for the pattern (EDA)	insight + correlation
5 · Model	Learn the temperature→sales relationship	a trained model
6 · Evaluate & Deploy	Test it, then predict & decide	a usable forecast

🎯 The project: "How many ice creams will we sell tomorrow?"¶

⚙️ Setup, imports & the book's plotting style¶

	temperature	sales
0	18	22.0
1	20	28.0
2	21	30.0
3	23	35.0
4	25	40.0
5	26	44.0
6	28	52.0
7	30	60.0
8	31	5.0
9	33	70.0
10	25	40.0
11	27	NaN
12	22	26.0
13	24	34.0