import gradio as gr import numpy as np import pandas as pd import matplotlib.pyplot as plt from sklearn import datasets from sklearn.preprocessing import StandardScaler from sklearn.decomposition import PCA import plotly.graph_objects as go import plotly.express as px import time FEATURE_LABELS = { "age": "Leeftijd", "sex": "Geslacht", "bmi": "BMI (Body Mass Index)", "bp": "Bloeddruk", "s1": "Totale cholesterol", "s2": "LDL-cholesterol", "s3": "HDL-cholesterol", "s4": "Chol./HDL-verhouding", "s5": "Triglyceriden", "s6": "Bloedsuiker (glucose)", "target": "Doelscore (progressie)", } LABEL_TO_KEY = {v: k for k, v in FEATURE_LABELS.items()} MEDICAL_MD = """ ### Medisch nut **Wat zien we hier?** Ik heb een bestaande, anonieme gezondheidsdataset gebruikt die speciaal beschikbaar is gemaakt voor onderzoek en studie. In deze gegevens staan metingen van een grote groep patiënten, zoals **bloedwaarden, BMI, cholesterol en bloedsuiker**. Zo'n enorme berg cijfers is voor artsen en ziekenhuizen bijna niet in één keer te overzien. Het is gewoon te veel om met het blote oog patronen uit te halen. **Daar komt kunstmatige intelligentie om de hoek kijken.** Met deze techniek (PCA) kan de computer de data slim samenvatten en patronen zichtbaar maken. Dit programma dat ik heb ontworpen laat live zien hoe die samenvatting werkt. - Elke punt is één patiënt. - De kleur laat zien hoe hoog of laag een bepaalde meting is (standaard: BMI). - De pijlen (in de 2D-biplot) laten zien welke metingen het meeste invloed hebben. - Links bovenin kun je kiezen welke meting je als uitgangspunt wilt nemen. **En wat heb je hieraan?** In de praktijk gebruiken artsen en onderzoekers zo'n plot om patronen en verbanden te ontdekken. 👉 Het is dus niet alleen een mooi plaatje, maar echt een manier om grote hoeveelheden data sneller en slimmer te begrijpen. Met AI kunnen we patronen vinden die je met het blote oog nooit zou zien. Dat maakt dit niet alleen een mooie visualisatie, maar ook een knap stukje technologie met échte waarde voor onderzoek en zorg. **Speel zelf de onderzoeker!** Doe alsof je een arts bent en kies links bovenin een waarde, bijvoorbeeld **cholesterol**, **leeftijd** of **geslacht**. Klik daarna op **Update visualisaties** en ontdek je eigen patronen in de data. """ # -------------------- Data helpers -------------------- def load_diabetes_df(): d = datasets.load_diabetes() X = pd.DataFrame(d.data, columns=d.feature_names) # gestandaardiseerd y = pd.Series(d.target, name="target") df = X.copy(); df["target"] = y return df def compute_overview_table(df: pd.DataFrame): keys = ["bmi","bp","s1","s2","s3","s4","s5","s6"] rows = [] for k in keys: vals = df[k].dropna().values mean = float(vals.mean()) pct_above = float((vals > 0).mean() * 100.0) # 0 ≈ globaal gemiddelde pct_below = float((vals < 0).mean() * 100.0) rows.append({ "Meting": FEATURE_LABELS.get(k, k), "Gemiddelde (gestandaardiseerd)": round(mean, 3), "% boven gemiddelde": round(pct_above, 1), "% onder gemiddelde": round(pct_below, 1), }) table = pd.DataFrame(rows) note = ("Let op: waarden in deze dataset zijn **gestandaardiseerd**. `0` ≈ algemeen gemiddelde. " "Positief = hoger dan gemiddeld, negatief = lager dan gemiddeld.") return table, note # -------------------- PCA helpers -------------------- def compute_pca(df: pd.DataFrame, n_components: int, standardize: bool): feats = [c for c in df.columns if c != "target"] X = df[feats].values if standardize: scaler = StandardScaler(with_mean=True, with_std=True) Xs = scaler.fit_transform(X) else: Xs = X pca = PCA(n_components=min(int(n_components), Xs.shape[1])) Z = pca.fit_transform(Xs) loadings = pca.components_.T expl = pca.explained_variance_ratio_ return feats, Xs, Z, loadings, expl # -------------------- Plot builders -------------------- def build_biplot_plotly(df, Z, loadings, feats, color_key, arrow_scale=2.0): # Hover info fields = ["bmi","bp","s1","s2","s3","s4","s5","s6","age","sex","target"] hover_text = [ "
".join(f"{FEATURE_LABELS.get(k,k)}: {df.iloc[i][k]:.3f}" for k in fields) for i in range(len(df)) ] fig = go.Figure() fig.add_trace(go.Scatter( x=Z[:,0], y=Z[:,1], mode="markers", marker=dict(size=8, color=df[color_key].values), text=hover_text, hovertemplate="%{text}" )) # loading pijlen for i, key in enumerate(feats): x = loadings[i,0]*arrow_scale; y = loadings[i,1]*arrow_scale fig.add_annotation(x=x, y=y, ax=0, ay=0, xref="x", yref="y", axref="x", ayref="y", showarrow=True, arrowhead=3) fig.add_annotation(x=x*1.05, y=y*1.05, text=FEATURE_LABELS.get(key,key), showarrow=False, font=dict(size=10)) fig.update_layout(title="PCA-biplot (2D, hover)", xaxis_title="PC1", yaxis_title="PC2", margin=dict(l=10, r=10, t=40, b=10)) return fig def build_biplot_matplotlib(df, Z, loadings, feats, color_key, arrow_scale=2.0, point_size=32, alpha=0.85): fig = plt.figure() ax = fig.add_subplot(111) sc = ax.scatter(Z[:,0], Z[:,1], c=df[color_key].values, s=point_size, alpha=alpha) cbar = plt.colorbar(sc, ax=ax, pad=0.02); cbar.set_label(f"Kleur: {FEATURE_LABELS.get(color_key,color_key)}") ax.set_xlabel("PC1"); ax.set_ylabel("PC2"); ax.set_title("PCA-biplot — PNG-export") for i,key in enumerate(feats): x=loadings[i,0]*arrow_scale; y=loadings[i,1]*arrow_scale ax.arrow(0,0,x,y, head_width=0.05, head_length=0.08, fc="k", ec="k", length_includes_head=True) ax.text(x*1.08, y*1.08, FEATURE_LABELS.get(key,key), fontsize=9, ha="center", va="center") ax.axhline(0,color="grey",linewidth=0.6,linestyle=":"); ax.axvline(0,color="grey",linewidth=0.6,linestyle=":") ax.grid(True,linestyle=":",linewidth=0.6); fig.tight_layout() return fig def build_pca3d(Z3, color_vals): fig = go.Figure(data=[go.Scatter3d(x=Z3[:,0], y=Z3[:,1], z=Z3[:,2], mode="markers", marker=dict(size=4, color=color_vals, opacity=0.85))]) fig.update_layout(title="PCA 3D — PC1·PC2·PC3 (sleep om te draaien)", scene=dict(xaxis_title="PC1", yaxis_title="PC2", zaxis_title="PC3"), margin=dict(l=10, r=10, t=40, b=10)) return fig def build_variance_plot(expl): fig = plt.figure() ax = fig.add_subplot(111) xs = np.arange(1, len(expl)+1) ax.bar(xs, expl, width=0.8, align="center") ax.plot(xs, np.cumsum(expl), marker="o") ax.set_xticks(xs); ax.set_xlabel("Principal Component"); ax.set_ylabel("Explained variance ratio") ax.set_title("Uitlegvariantie per component (balken) + cumulatief (lijn)") ax.grid(True, linestyle=":", linewidth=0.6); fig.tight_layout() return fig def build_hist_box(df: pd.DataFrame, color_key: str): series = df[color_key].dropna() label = FEATURE_LABELS.get(color_key, color_key) fig_hist = px.histogram(x=series, nbins=30, title=f"Histogram — {label}", labels={"x": label}) fig_hist.update_layout(xaxis_title=label, yaxis_title="Aantal", margin=dict(l=10, r=10, t=40, b=10)) fig_box = px.box(y=series, points="outliers", title=f"Boxplot — {label}", labels={"y": label}) fig_box.update_layout(yaxis_title=label, margin=dict(l=10, r=10, t=40, b=10)) return fig_hist, fig_box # -------------------- Controllers -------------------- def controller(color_label="BMI (Body Mass Index)", n_components=10, standardize=True, arrow_scale=2.0): df = load_diabetes_df() feats, Xs, Z, loadings, expl = compute_pca(df, n_components, standardize) color_key = LABEL_TO_KEY.get(color_label, "bmi") color_vals = df[color_key].values fig_biplot = build_biplot_plotly(df, Z, loadings, feats, color_key, arrow_scale=arrow_scale) if Z.shape[1] < 3: pca3 = PCA(n_components=3); Z3 = pca3.fit_transform(Xs) else: Z3 = Z[:, :3] fig3d = build_pca3d(Z3, color_vals) fig_variance = build_variance_plot(expl) fig_hist, fig_box = build_hist_box(df, color_key) load_df = pd.DataFrame({ "feature_key": feats, "PC1_loading": loadings[:, 0], "PC2_loading": loadings[:, 1], "PC1_abs": np.abs(loadings[:, 0]), "PC2_abs": np.abs(loadings[:, 1]), }) load_df["Feature (PC1)"] = load_df["feature_key"].map(lambda k: FEATURE_LABELS.get(k, k)) load_df["Feature (PC2)"] = load_df["feature_key"].map(lambda k: FEATURE_LABELS.get(k, k)) top_pc1 = load_df.sort_values("PC1_abs", ascending=False)[["Feature (PC1)", "PC1_loading"]].head(6).reset_index(drop=True) top_pc2 = load_df.sort_values("PC2_abs", ascending=False)[["Feature (PC2)", "PC2_loading"]].head(6).reset_index(drop=True) max_len = max(len(top_pc1), len(top_pc2)) top_pc1 = top_pc1.reindex(range(max_len)); top_pc2 = top_pc2.reindex(range(max_len)) table = pd.concat([top_pc1, top_pc2], axis=1) overview_df, overview_note = compute_overview_table(df) summary_md = f""" ### Wat zie je hier? - **Klik op _Update visualisaties_** om alles te verversen met jouw keuze. - **Hover** over punten voor exacte waarden (BMI, bloeddruk, cholesterol, glucose, leeftijd, geslacht, etc.). - **2D-biplot** met pijlen (belangrijkste metingen) en **3D-view** voor extra diepte. - **Uitlegvariantieplot**: laat zien hoeveel variatie elke component uitlegt. - **Histogram + boxplot**: verdeling en spreiding van de gekozen meting ({FEATURE_LABELS.get(color_key,color_key)}). """ return fig_biplot, fig3d, fig_variance, table, overview_df, overview_note, summary_md def animate_pca(color_label="BMI (Body Mass Index)", point_size=32, alpha=0.85, n_components=10, standardize=True, frames=40, pause=0.0): df = load_diabetes_df() feats, Xs, Z, loadings, expl = compute_pca(df, n_components, standardize) color_key = LABEL_TO_KEY.get(color_label, "bmi") color_vals = df[color_key].values for i in range(frames): t = i / max(1, frames-1) w1 = min(1.0, t * 2.0); w2 = max(0.0, (t - 0.5) * 2.0) coords = np.column_stack([Z[:, 0] * w1, Z[:, 1] * w2]) fig = plt.figure() ax = fig.add_subplot(111) ax.scatter(coords[:, 0], coords[:, 1], c=color_vals, s=point_size, alpha=alpha) ax.set_xlabel("PC1 (opbouw)"); ax.set_ylabel("PC2 (opbouw)") title = "PCA-projectie (animatie) — " + ("PC1 →" if w2 == 0 else "PC1 + PC2") ax.set_title(f"{title} — frame {i+1}/{frames}") ax.axhline(0, color="grey", linewidth=0.6, linestyle=":"); ax.axvline(0, color="grey", linewidth=0.6, linestyle=":") ax.grid(True, linestyle=":", linewidth=0.6); fig.tight_layout() yield fig if pause > 0: time.sleep(pause) def export_biplot_png(color_label="BMI (Body Mass Index)", arrow_scale=2.0, point_size=32, alpha=0.85, n_components=10, standardize=True): df = load_diabetes_df() feats, Xs, Z, loadings, expl = compute_pca(df, n_components, standardize) color_key = LABEL_TO_KEY.get(color_label, "bmi") fig = build_biplot_matplotlib(df, Z, loadings, feats, color_key, arrow_scale=arrow_scale, point_size=point_size, alpha=alpha) path = f"/mnt/data/biplot_{int(time.time())}.png" fig.savefig(path, dpi=150, bbox_inches="tight"); plt.close(fig) return path def export_variance_png(n_components=10, standardize=True): df = load_diabetes_df() feats, Xs, Z, loadings, expl = compute_pca(df, n_components, standardize) fig = build_variance_plot(expl) path = f"/mnt/data/variance_{int(time.time())}.png" fig.savefig(path, dpi=150, bbox_inches="tight"); plt.close(fig) return path # -------------------- UI -------------------- with gr.Blocks(title="PCA Dashboard — Diabetes (netjes & compleet)") as demo: gr.HTML(""" """) gr.Markdown("# PCA Dashboard — Diabetes (netjes & compleet)") gr.Markdown(MEDICAL_MD) gr.HTML('
Belangrijk: kies links je instellingen en klik daarna op Update visualisaties. Wil je de stap-voor-stap projectie zien? Klik op ▶ Animate PCA.
') with gr.Row(): with gr.Column(scale=1): with gr.Group(): gr.Markdown("### Instellingen") color_choices = [FEATURE_LABELS[k] for k in ["bmi","bp","s1","s2","s3","s4","s5","s6","age","sex","target"]] color_feat = gr.Dropdown(choices=color_choices, value=FEATURE_LABELS["bmi"], label="Kleur op meting") n_components = gr.Slider(3, 10, value=10, step=1, label="Aantal PCA-componenten") standardize = gr.Checkbox(value=True, label="Standaardiseer metingen (aanbevolen)") arrow_scale = gr.Slider(0.5, 5.0, value=2.0, step=0.1, label="Pijl-schaal (2D-biplot)") run_btn = gr.Button("🔄 Update visualisaties") gr.HTML('
Klik hierna op: "🔄 Update visualisaties" om alle grafieken te verversen.
') with gr.Group(): gr.Markdown("### Animatie") animate_btn = gr.Button("▶ Animate PCA (PC1 → PC2)") gr.HTML('
Klik op: "▶ Animate PCA" om de projectie stap-voor-stap te zien.
') anim_plot = gr.Plot(label="Animatie van projectie") with gr.Group(): gr.Markdown("### Downloads") dl_biplot = gr.DownloadButton("Download biplot (PNG)") dl_var = gr.DownloadButton("Download variatieplot (PNG)") with gr.Column(scale=2): with gr.Row(): with gr.Column(): gr.Markdown("### Biplot (2D, hover)") plot_biplot = gr.Plotly() with gr.Column(): gr.Markdown("### 3D PCA (PC1–PC3)") plot3d = gr.Plotly() with gr.Row(): with gr.Column(): gr.Markdown("### Uitlegvariantie") plot_expl = gr.Plot() with gr.Column(): gr.Markdown("### Top-features (PC1 / PC2)") table = gr.Dataframe(headers=["Feature (PC1)", "Loading PC1", "Feature (PC2)", "Loading PC2"], row_count=6) with gr.Row(): with gr.Column(): gr.Markdown("### Histogram") plot_hist = gr.Plotly() with gr.Column(): gr.Markdown("### Boxplot") plot_box = gr.Plotly() with gr.Row(): with gr.Column(): gr.Markdown("### Overzicht (gemiddelden & verdeling)") overview_tbl = gr.Dataframe(interactive=False) with gr.Column(): gr.Markdown("### Samenvatting") summary = gr.Markdown() overview_note_md = gr.Markdown() inputs = [color_feat, n_components, standardize, arrow_scale] run_btn.click(fn=controller, inputs=inputs, outputs=[plot_biplot, plot3d, plot_expl, table, overview_tbl, overview_note_md, summary]) demo.load(fn=controller, inputs=inputs, outputs=[plot_biplot, plot3d, plot_expl, table, overview_tbl, overview_note_md, summary]) animate_btn.click(fn=animate_pca, inputs=[color_feat], outputs=anim_plot) dl_biplot.click(fn=export_biplot_png, inputs=[color_feat, arrow_scale], outputs=[dl_biplot]) dl_var.click(fn=export_variance_png, inputs=[], outputs=[dl_var]) if __name__ == "__main__": demo.queue().launch(server_name="0.0.0.0", server_port=7860, ssr_mode=False, show_api=False)