Upload EDA.py

EDA.py

@@ -0,0 +1,90 @@
+import numpy as np
+import pandas as pd
+import matplotlib.pyplot as plt
+import seaborn as sns
+from pathlib import Path
+
+PATH = Path("BuildingsIncomePopulationDataset.csv")   
+OUT = Path("Visualization")     
+OUT.mkdir(parents=True, exist_ok=True)                                    
+sns.set_theme(style="whitegrid")
+
+# Load
+# Semicolon separated, decimal comma, postal codes keep leading zeros
+df = pd.read_csv(PATH, sep=";", decimal=",", encoding="utf-8-sig", dtype={"Postal code": str})
+df.columns = ["Postal_code", "Over_18", "Mean_income", "Household_size", "Buildings",
+              "Households", "Population", "Median_income", "Residential_buildings"]
+
+print("Dimensions:", df.shape)
+
+num = df.columns.drop("Postal_code")
+
+# Verify data
+print("Residential buildings > total buildings:", (df.Residential_buildings > df.Buildings).sum())
+print("Population / (Households * Household_size), should be ~1:\n",
+      (df["Population"] / (df.Households * df.Household_size)).describe().round(3), "\n")
+
+#Deprived data
+df["Adult_ratio"]   = df.Over_18 / df["Population"].replace(0, np.nan)
+df["Income_ratio"]  = df.Mean_income / df.Median_income          # >1 => right-skewed income
+df["Density"] = df["Population"] / df.Buildings                   # residents per building
+df["Residential_ratio"]   = df.Residential_buildings / df.Buildings
+
+print(df[num].describe().T.round(1), "\n")
+print("Skewness:\n", df[num].skew().round(2), "\n")
+
+
+core = ["Population", "Households", "Buildings", "Mean_income", "Median_income", "Household_size"]
+fig, axes = plt.subplots(2, 3, figsize=(14, 7))
+for ax, c in zip(axes.ravel(), core):
+    sns.histplot(df[c], kde=True, ax=ax, color="Steelblue")
+    ax.axvline(df[c].median(), color="crimson", ls="--", lw=1, label="Median")
+    ax.set_title(c); ax.set_xlabel("")
+axes[0, 0].legend()
+fig.suptitle("Distributions across postal code areas", y=1.0)
+fig.tight_layout(); fig.savefig(f"{OUT}/01_distributions.png", dpi=150); plt.close(fig)
+
+fig, axes = plt.subplots(1, len(core), figsize=(15, 3.5))
+for ax, c in zip(axes, core):
+    sns.boxplot(y=df[c], ax=ax, color="lightsteelblue"); ax.set_title(c); ax.set_ylabel("")
+fig.tight_layout(); fig.savefig(f"{OUT}/02_boxplots.png", dpi=150); plt.close(fig)
+
+# Outliers via 1.5*IQR rule, all columns at once
+q1, q3 = df[core].quantile([.25, .75]).values
+iqr = q3 - q1
+mask = (df[core] < q1 - 1.5 * iqr) | (df[core] > q3 + 1.5 * iqr)
+print("IQR outlier counts:\n", mask.sum(), "\n")
+print("Postal codes flagged as outliers:\n",
+      mask.apply(lambda s: df.loc[s, "Postal_code"].tolist()).to_string(), "\n")
+
+#Bivariate
+corr = df[core + ["Over_18", "Residential_buildings"]].corr(method="spearman")
+fig, ax = plt.subplots(figsize=(8, 6.5))
+sns.heatmap(corr, annot=True, fmt=".2f", cmap="coolwarm", center=0, vmin=-1, vmax=1, ax=ax)
+ax.set_title("Spearman correlation")
+fig.tight_layout(); fig.savefig(f"{OUT}/03_correlation.png", dpi=150); plt.close(fig)
+print("Strongest pairs (|rho|, upper triangle):\n",
+      corr.where(np.triu(np.ones(corr.shape, bool), 1)).stack().abs().nlargest(8).round(2), "\n")
+
+fig, axes = plt.subplots(1, 3, figsize=(16, 4.8))
+sns.scatterplot(data=df, x="Median_income", y="Mean_income", size="Population", sizes=(15, 250),
+                alpha=.6, ax=axes[0], legend=False)
+lim = [df.Median_income.min(), df.Median_income.max()]
+axes[0].plot(lim, lim, "k--", lw=1); axes[0].set_title("Mean vs median income (dashed: mean = median)")
+sns.regplot(data=df, x="Population", y="Households", ax=axes[1], scatter_kws=dict(alpha=.6))
+axes[1].set_title("Inhabitants vs households")
+sns.scatterplot(data=df, x="Population", y="Median_income", hue="Household_size", palette="viridis", ax=axes[2])
+axes[2].set_title("Inhabitants vs median income")
+fig.tight_layout(); fig.savefig(f"{OUT}/04_relationships.png", dpi=150); plt.close(fig)
+
+#Rankings and structure
+cols = ["Postal_code", "Population", "Mean_income", "Median_income", "Income_ratio"]
+print("Highest mean income:\n", df.nlargest(5, "Mean_income")[cols].round(2), "\n")
+print("Lowest median income:\n", df.nsmallest(5, "Median_income")[cols].round(2), "\n")
+print("Largest by population:\n", df.nlargest(5, "Population")[["Postal_code", "Population", "Households", "Buildings"]], "\n")
+
+# Income inequality: mean/median ratio vs mean income
+fig, ax = plt.subplots(figsize=(6.5, 5))
+sns.scatterplot(data=df, x="Mean_income", y="Income_ratio", ax=ax)
+ax.set_title("Income skew (mean/median) vs mean income")
+fig.tight_layout(); fig.savefig(f"{OUT}/05_income_skew.png", dpi=150); plt.close(fig)