add: AIDS dataset quantile regression

Author: vsheg

03-nonlinear/aids.py

@@ -0,0 +1,49 @@
+# %%
+import numpy as np
+import pandas as pd
+import seaborn as sns
+from sksurv.datasets import load_aids
+from sklearn.linear_model import QuantileRegressor
+import scienceplots
+import matplotlib.pyplot as plt
+
+# %%
+X, y_frame = load_aids()
+X = X.astype(float)
+
+# %%
+y_data = pd.DataFrame(y_frame)
+_, y = y_data.values.T
+y = y.astype(float)
+
+# %%
+quantiles = np.linspace(0.1, 0.9, 20)
+coeffs = {}
+
+for q in quantiles:
+    model = QuantileRegressor(quantile=q, alpha=0.01)
+    model.fit(X, y)
+    coeffs[q] = model.coef_
+
+# %%
+df_coeffs = pd.DataFrame(coeffs, index=X.columns)
+features_important = df_coeffs.abs().sum(axis=1).sort_values(ascending=False).head(4)
+df_coeffs = df_coeffs.loc[features_important.index].reset_index(names="feature")
+df_coeffs = df_coeffs.melt("feature", var_name="quantile", value_name="coefficient")
+
+# %%
+plt.style.use("default")
+plt.style.use("no-latex")
+plot = sns.relplot(
+    data=df_coeffs,
+    x="quantile",
+    y="coefficient",
+    col="feature",
+    col_wrap=2,
+    kind="line",
+    height=2,
+    aspect=4 / 3,
+)
+
+# %%
+plot.savefig("aids.svg")