""" ╔══════════════════════════════════════════════════════════════════════════════╗ ║ CopaVision AI | Phase 2: Player Performance Intelligence ║ ║ Rebuilt for Beginner-Friendly UX — v2 ║ ╚══════════════════════════════════════════════════════════════════════════════╝ Save as: pages/02_Player_Dashboard.py Data files required (in data/ folder next to app.py): data/player_stats.csv data/player_per90.csv data/similarity_features.csv data/clustered_player_features.csv """ import warnings warnings.filterwarnings("ignore") from pathlib import Path import numpy as np import pandas as pd import plotly.graph_objects as go import plotly.express as px from sklearn.cluster import KMeans from sklearn.decomposition import PCA from sklearn.manifold import TSNE from sklearn.metrics.pairwise import cosine_similarity import streamlit as st # ───────────────────────────────────────────────────────────────────────────── # PAGE CONFIG # ───────────────────────────────────────────────────────────────────────────── st.set_page_config( page_title="Player Intelligence | CopaVision AI", page_icon="🧠", layout="wide", initial_sidebar_state="expanded", ) # ───────────────────────────────────────────────────────────────────────────── # CSS — same dark football theme as Phase 1 + new helper classes # ───────────────────────────────────────────────────────────────────────────── st.markdown(""" """, unsafe_allow_html=True) # ───────────────────────────────────────────────────────────────────────────── # CONSTANTS # ───────────────────────────────────────────────────────────────────────────── DATA_DIR = Path(__file__).parent.parent / "data" MIN_MINUTES = 450 C = { # colour palette shorthand "accent": "#00d4ff", "green": "#00ff9d", "orange": "#ff6b35", "gold": "#ffd700", "purple": "#d2a8ff", "muted": "#5a7a9a", "bg": "#0e1621", "border": "#1e3050", "text": "#e8f1ff", } POSITION_GROUPS = { "⛳ Goalkeeper": ["Goalkeeper"], "🛡️ Defender": ["Center Back","Left Back","Right Back","Left Center Back", "Right Center Back","Left Wing Back","Right Wing Back"], "⚙️ Midfielder": ["Center Midfield","Left Center Midfield","Right Center Midfield", "Center Defensive Midfield","Left Defensive Midfield", "Right Defensive Midfield","Left Midfield","Right Midfield"], "🌪️ Attacking Mid / Winger": ["Center Attacking Midfield","Left Attacking Midfield", "Right Attacking Midfield","Left Wing","Right Wing"], "⚡ Forward": ["Center Forward","Left Center Forward","Right Center Forward", "Secondary Striker"], } # Plain-English position group names (strip emoji for internal logic) def _pos_group_key(label): return label.split(" ", 1)[1] if " " in label else label CLUSTER_ARCHETYPES = { 0: {"name":"Defensive Anchor", "color":"#ff6b35","icon":"🛡️", "desc":"Wins headers, blocks shots, makes last-ditch tackles. The rock at the back."}, 1: {"name":"Ball Winner", "color":"#d2a8ff","icon":"⚔️", "desc":"Hunts the ball relentlessly. High in duels, interceptions, and pressing intensity."}, 2: {"name":"Deep-Lying Playmaker", "color":"#00d4ff","icon":"🎯", "desc":"The conductor. Passes the most, most accurately, and moves the ball forward."}, 3: {"name":"Clinical Finisher", "color":"#ffd700","icon":"⚡", "desc":"Lives in the box. High xG, shots on target, and lethal conversion rate."}, 4: {"name":"Box-to-Box Engine", "color":"#00ff9d","icon":"🔄", "desc":"Covers every blade of grass. Contributes in both attack and defence equally."}, 5: {"name":"Creative Winger", "color":"#ff6b35","icon":"🌪️", "desc":"Takes on defenders 1v1, creates from wide areas, and delivers assists."}, 6: {"name":"High Press Striker", "color":"#ffd700","icon":"🎯", "desc":"Scores goals AND pressures defenders. A nightmare to play against."}, 7: {"name":"Possession Carrier", "color":"#00d4ff","icon":"🎪", "desc":"Drives forward with the ball, beats players, and progresses play."}, } FEATURE_COLS_SIM = [ "goals_per90_scaled","assists_per90_scaled","shots_per90_scaled", "shots_on_target_per90_scaled","xg_per90_scaled", "passes_attempted_per90_scaled","passes_completed_per90_scaled", "progressive_passes_per90_scaled","dribbles_attempted_per90_scaled", "dribbles_completed_per90_scaled","carries_per90_scaled", "tackles_per90_scaled","tackles_won_per90_scaled","interceptions_per90_scaled", "pressures_per90_scaled","aerial_duels_per90_scaled","aerial_duels_won_per90_scaled", "duels_per90_scaled","duels_won_per90_scaled","pass_accuracy_scaled", "dribble_success_rate_scaled","tackle_success_rate_scaled", "aerial_success_rate_scaled","duel_success_rate_scaled", ] # ───────────────────────────────────────────────────────────────────────────── # DATA LOADING # ───────────────────────────────────────────────────────────────────────────── @st.cache_data(show_spinner=False) def load_player_stats(): df = pd.read_csv(DATA_DIR / "player_stats.csv") nickname = df["player_nickname"].replace(["0", 0, "", "nan", "None"], np.nan) df["display_name"] = nickname.fillna(df["player_name"]) df = df[df["position"] != "0"].copy() # Attach readable position group label def _group(pos): for label, positions in POSITION_GROUPS.items(): if pos in positions: return label return "Other" df["pos_group"] = df["position"].apply(_group) return df @st.cache_data(show_spinner=False) def load_per90(): df = pd.read_csv(DATA_DIR / "player_per90.csv") nickname = df["player_nickname"].replace(["0", 0, "", "nan", "None"], np.nan) df["display_name"] = nickname.fillna(df["player_name"]) df = df[df["position"] != "0"].copy() return df @st.cache_data(show_spinner=False) def load_similarity(): df = pd.read_csv(DATA_DIR / "similarity_features.csv").fillna(0) display_lookup = load_player_stats()[["player_id", "display_name"]].drop_duplicates("player_id") df = df.merge(display_lookup, on="player_id", how="left") df["display_name"] = df["display_name"].fillna(df["player_name"]) return df @st.cache_data(show_spinner=False) def load_clusters(): REQUIRED = {"player_id","cluster","pca_x","pca_y","tsne_x","tsne_y"} path = DATA_DIR / "clustered_player_features.csv" if path.exists(): df = pd.read_csv(path) if REQUIRED.issubset(set(df.columns)): display_lookup = load_player_stats()[["player_id", "display_name"]].drop_duplicates("player_id") df = df.merge(display_lookup, on="player_id", how="left") df["display_name"] = df["display_name"].fillna(df["player_name"]) return df return _build_clusters() @st.cache_data(show_spinner=False) def _build_clusters(): sim = load_similarity() X = sim[FEATURE_COLS_SIM].values cl = sim[["player_id","player_name","primary_position","team_name","competition_name"]].copy() cl["cluster"] = KMeans(n_clusters=8, random_state=42, n_init=10).fit_predict(X) pca = PCA(n_components=2, random_state=42).fit_transform(X) tsne = TSNE(n_components=2, random_state=42, perplexity=30).fit_transform(X) cl["pca_x"], cl["pca_y"] = pca[:,0], pca[:,1] cl["tsne_x"], cl["tsne_y"] = tsne[:,0], tsne[:,1] display_lookup = load_player_stats()[["player_id", "display_name"]].drop_duplicates("player_id") cl = cl.merge(display_lookup, on="player_id", how="left") cl["display_name"] = cl["display_name"].fillna(cl["player_name"]) try: cl.to_csv(DATA_DIR / "clustered_player_features.csv", index=False) except Exception: pass return cl # ───────────────────────────────────────────────────────────────────────────── # SMALL HELPERS # ───────────────────────────────────────────────────────────────────────────── def fmt(val, dec=2, suffix=""): if pd.isna(val): return "—" if isinstance(val, float): return f"{val:.{dec}f}{suffix}" return f"{val}{suffix}" def pct_rank(series, val): s = series.dropna() if s.std() == 0: return 50.0 return float((s < val).mean() * 100) def pct_color(p): if p >= 85: return C["green"] if p >= 65: return C["accent"] if p >= 40: return C["gold"] return C["orange"] def pct_bar(label, pct, raw="", explain=""): col = pct_color(pct) badge = ("🔥 Elite" if pct>=85 else "✅ Good" if pct>=65 else "➡️ Average" if pct>=40 else "📉 Below Avg") tip = f" · {explain}" if explain else "" return f"""
{label}{tip} {raw} {pct:.0f}th · {badge}
""" def stat_row(label, val, color=None, tooltip=""): color = color or C["text"] return (f'
' f'{label}' f'{val}' f'
') def _PLOT_BASE(): return dict(template="plotly_dark", paper_bgcolor="rgba(0,0,0,0)", plot_bgcolor="rgba(0,0,0,0)") # ───────────────────────────────────────────────────────────────────────────── # GLOSSARY COMPONENT (shown inline wherever metrics appear) # ───────────────────────────────────────────────────────────────────────────── GLOSSARY = { "xG (Expected Goals)": "A probability score (0–1) for each shot — how likely it was to go in based on position, angle, and body part. " "Higher xG = better quality chances. A striker with xG 20 but only 10 goals is under-performing; 25 goals = over-performing.", "xA (Expected Assists)": "Same idea as xG but for passes that led to shots. Measures the quality of chances created, not just whether the finish went in.", "Per 90 (p90)": "Stats divided by 90 minutes played, so you can fairly compare a player who played 900 mins to one who played 3,600 mins.", "Percentile Rank": "Your rank vs ALL players in the database. 90th percentile = better than 90% of players at that stat. 50th = exactly average.", "Progressive Passes": "Passes that move the ball at least 10 metres closer to the opponent's goal. Measures forward intent in build-up play.", "Pressures": "How often the player actively tries to win the ball back by pressuring the opponent in possession. A key pressing metric.", "Duel Success Rate": "% of 1v1 physical battles won. High rate = dominant in physical contests.", } def glossary_expander(): with st.expander("📖 Glossary — What do these numbers mean?", expanded=False): for term, definition in GLOSSARY.items(): st.markdown(f"""
{term}
{definition}
""", unsafe_allow_html=True) # ───────────────────────────────────────────────────────────────────────────── # SIDEBAR # ───────────────────────────────────────────────────────────────────────────── def render_sidebar(): with st.sidebar: # Brand st.markdown("""
COPAVISION
A I · P H A S E 2

""", unsafe_allow_html=True) # Navigation with plain labels st.markdown('CHOOSE A TOOL', unsafe_allow_html=True) page = st.radio("nav", [ "🔍 Search a Player", "⚖️ Compare Two Players", "🗺️ Explore Player Types", "🔗 Find Similar Players", ], label_visibility="collapsed") st.markdown("
", unsafe_allow_html=True) st.markdown("""
Choose a tool above, then use the page controls to filter players and compare stats.
""", unsafe_allow_html=True) # Strip emoji prefix for routing return page.strip().split(" ")[-1].strip() # ───────────────────────────────────────────────────────────────────────────── # ── SHARED: STEP-BY-STEP PLAYER PICKER (used on 3 pages) # ───────────────────────────────────────────────────────────────────────────── def player_picker(ps_df, pp_df, key_prefix="pp", min_mins=MIN_MINUTES, show_howto=True, label_override=None): """ Guided 3-step player picker: Step 1 — choose position group Step 2 — choose competition (optionally also team) Step 3 — choose player from the filtered shortlist Returns: (player_row from ps_df, per90_row from pp_df, player_id) """ # ── Onboarding banner ───────────────────────────────────────────────────── if show_howto: st.markdown("""
🧭 HOW TO PICK A PLAYER — 3 EASY STEPS
1
Choose a position group — are you looking for a goalkeeper, defender, midfielder, or forward? Leave it on "All positions" to browse everyone.
2
Pick a competition (league/tournament) and optionally a team. This narrows the player list so it's not overwhelming.
3
Select the player from the dropdown. Stats load instantly below.
""", unsafe_allow_html=True) # ── STEP 1: Position ────────────────────────────────────────────────────── st.markdown("""
1
Choose a position group
""", unsafe_allow_html=True) pos_options = ["All positions"] + list(POSITION_GROUPS.keys()) pos_labels = {opt: opt for opt in pos_options} pos_choice = st.selectbox( "position group", pos_options, index=0, format_func=lambda x: x, label_visibility="collapsed", key=f"{key_prefix}_pos", help="Pick the type of player you want to explore. Select 'All positions' to see every player.", ) # Filter by position if pos_choice == "All positions": filtered = ps_df.copy() else: allowed = POSITION_GROUPS.get(pos_choice, []) filtered = ps_df[ps_df["position"].isin(allowed)].copy() st.markdown("
", unsafe_allow_html=True) # ── STEP 2: Competition + Team ───────────────────────────────────────────── st.markdown("""
2
Narrow by competition, team & nationality (optional)
""", unsafe_allow_html=True) col_comp, col_team, col_country = st.columns(3) with col_comp: comps = ["All competitions"] + sorted(filtered["competition"].unique()) comp_choice = st.selectbox( "competition", comps, index=0, label_visibility="collapsed", key=f"{key_prefix}_comp", help="e.g. 'Premier League', 'FIFA World Cup', 'La Liga'", ) if comp_choice != "All competitions": filtered = filtered[filtered["competition"] == comp_choice] with col_team: teams = ["All teams"] + sorted(filtered["team"].unique()) team_choice = st.selectbox( "team", teams, index=0, label_visibility="collapsed", key=f"{key_prefix}_team", help="Filter to one club or national team", ) if team_choice != "All teams": filtered = filtered[filtered["team"] == team_choice] with col_country: countries = ["All nationalities"] + sorted(filtered["country"].dropna().astype(str).unique()) country_choice = st.selectbox( "nationality", countries, index=0, label_visibility="collapsed", key=f"{key_prefix}_country", help="Filter by player nationality when available", ) if country_choice != "All nationalities": filtered = filtered[filtered["country"].astype(str) == country_choice] if "age" in filtered.columns and filtered["age"].notna().any(): min_age = int(np.nanmin(filtered["age"])) max_age = int(np.nanmax(filtered["age"])) if min_age < max_age: age_range = st.slider( "Age range", min_age, max_age, (min_age, max_age), key=f"{key_prefix}_age", help="Only shown when the processed dataset includes an age column.", ) filtered = filtered[filtered["age"].between(age_range[0], age_range[1])] # Minimum minutes guard filtered = filtered[filtered["minutes_played"] >= min_mins] # How many players remain n_avail = filtered["display_name"].nunique() if n_avail == 0: st.warning("⚠️ No players match these filters. Try choosing 'All competitions' or a different position.") return None, None, None st.markdown(f"""
✅ {n_avail} players available with ≥{min_mins} minutes played
""", unsafe_allow_html=True) # ── STEP 3: Player ───────────────────────────────────────────────────────── label = label_override or "3" st.markdown(f"""
{label}
Select the player
""", unsafe_allow_html=True) player_names = sorted(filtered["display_name"].unique()) selected = st.selectbox( "player", player_names, label_visibility="collapsed", key=f"{key_prefix}_player", help="Type a name to search, or scroll through the list.", ) # Resolve rows ps_rows = filtered[filtered["display_name"] == selected] player_row = ps_rows.sort_values("minutes_played", ascending=False).iloc[0] pid = player_row["player_id"] pp_rows = pp_df[pp_df["player_id"] == pid] if pp_rows.empty: pp_rows = pp_df[pp_df["display_name"] == selected] per90_row = (pp_rows.sort_values("minutes_played", ascending=False).iloc[0] if not pp_rows.empty else pd.Series(dtype=float)) return player_row, per90_row, int(pid) # ───────────────────────────────────────────────────────────────────────────── # PROFILE HEADER CARD # ───────────────────────────────────────────────────────────────────────────── def render_profile_header(player_row, cl_df): arch = {} if cl_df is not None: cl_rows = cl_df[cl_df["player_id"] == player_row["player_id"]] if not cl_rows.empty: try: arch = CLUSTER_ARCHETYPES.get(int(cl_rows.iloc[0]["cluster"]), {}) except Exception: arch = {} ac = arch.get("color", C["accent"]) an = arch.get("name", "Unknown") ai = arch.get("icon", "⚽") ad = arch.get("desc", "") mins = player_row.get("minutes_played", 0) match = player_row.get("matches_played", 0) comp = player_row.get("competition", "—") team = player_row.get("team", "—") pos = player_row.get("position", "—") nat = player_row.get("country", "—") name = player_row.get("display_name", player_row.get("player_name","?")) st.markdown(f"""
{name}
🏟️ {team}  ·  📍 {pos}  ·  🌍 {nat}  ·  🏆 {comp}
⏱️ {fmt(mins,0)} minutes played across {fmt(match,0)} matches
{ai} {an}
{ad}
""", unsafe_allow_html=True) # ───────────────────────────────────────────────────────────────────────────── # PERCENTILE BAR CHART (Plotly — beginner friendly labels) # ───────────────────────────────────────────────────────────────────────────── def make_pct_chart(per90_row, pp_df, title=""): METRICS = [ ("Goals per 90 min", "goals_per90", "Goals scored per 90 min"), ("Assists per 90 min", "assists_per90", "Goal assists per 90 min"), ("Expected Goals (xG) / 90", "xg_per90", "Shot quality score / 90"), ("Expected Assists (xA) / 90", "xassists_per90", "Chance creation quality"), ("Progressive Passes / 90", "progressive_passes_per90", "Forward-moving passes / 90"), ("Completed Dribbles / 90", "dribbles_completed_per90", "Dribbles won / 90"), ("Tackles / 90", "tackles_per90", "Tackles attempted / 90"), ("Interceptions / 90", "interceptions_per90", "Passes cut out / 90"), ("Pressures / 90", "pressures_per90", "Press attempts / 90"), ("Pass Accuracy %", "pass_accuracy", "% passes reaching teammate"), ] labels, pcts, colors, raws = [], [], [], [] for lbl, col, _ in METRICS: if col in pp_df.columns and col in per90_row.index: p = pct_rank(pp_df[col], per90_row.get(col, 0)) labels.append(lbl); pcts.append(p) colors.append(pct_color(p)) raws.append(fmt(per90_row.get(col, 0))) fig = go.Figure(go.Bar( x=pcts, y=labels, orientation="h", marker=dict(color=colors, line=dict(width=0)), text=[f"{p:.0f}th percentile" for p in pcts], textposition="outside", textfont=dict(family="Barlow Condensed", size=11, color=C["text"]), customdata=raws, hovertemplate="%{y}
Value: %{customdata}
Percentile: %{x:.0f}th", )) # Add a vertical "Average" line at 50th percentile fig.add_vline(x=50, line=dict(color=C["muted"], dash="dash", width=1), annotation_text="Average (50th)", annotation_position="top", annotation_font=dict(color=C["muted"], size=10)) fig.update_layout( **_PLOT_BASE(), xaxis=dict(range=[0,118], showgrid=False, visible=False), yaxis=dict(tickfont=dict(family="DM Sans", size=11, color=C["text"])), height=400, margin=dict(t=30,b=10,l=10,r=80), title=dict(text=title or "How does this player rank vs all players?", font=dict(family="Barlow Condensed", size=13, color=C["text"])), ) return fig # ───────────────────────────────────────────────────────────────────────────── # RADAR CHART # ───────────────────────────────────────────────────────────────────────────── RADAR_CATS = { "⚽ Finishing": ["goals_per90","xg_per90","shot_accuracy"], "🎨 Creativity": ["assists_per90","xassists_per90","progressive_passes_per90"], "🚗 Ball Carrying": ["dribbles_completed_per90","carries_per90"], "🛡️ Defending": ["tackles_per90","interceptions_per90","clearances_per90"], "💪 Physicality": ["aerial_duels_per90","duels_won_per90"], "📨 Passing": ["passes_completed_per90","pass_accuracy"], } def make_radar(r1, r2, n1, n2, pp_df): labels, v1, v2 = [], [], [] for cat, cols in RADAR_CATS.items(): for col in cols: if col in pp_df.columns: labels.append( f"{cat}\n{col.replace('_per90','').replace('_',' ').title()}" ) v1.append(pct_rank(pp_df[col], r1.get(col, 0))) v2.append(pct_rank(pp_df[col], r2.get(col, 0))) fig = go.Figure() for nm, vals, col, fill_col in [ (n1, v1, C["green"], "rgba(0,255,157,0.20)"), (n2, v2, C["accent"], "rgba(0,212,255,0.15)") ]: fig.add_trace(go.Scatterpolar( r=vals + [vals[0]], theta=labels + [labels[0]], fill="toself", fillcolor=fill_col, line=dict(color=col, width=2), name=nm, hovertemplate="%{theta}
Percentile: %{r:.0f}", )) fig.update_layout( **_PLOT_BASE(), polar=dict( bgcolor="rgba(14,22,33,0.6)", radialaxis=dict( visible=True, range=[0,100], showticklabels=False, gridcolor=C["border"] ), angularaxis=dict( tickfont=dict( family="DM Sans", size=9, color=C["text"] ), gridcolor=C["border"] ), ), legend=dict( font=dict( family="Barlow Condensed", color=C["text"], size=13 ), bgcolor="rgba(0,0,0,0)" ), height=420, margin=dict(t=40,b=40,l=50,r=50), ) return fig # ───────────────────────────────────────────────────────────────────────────── # CLUSTER SCATTER # ───────────────────────────────────────────────────────────────────────────── def make_cluster_scatter(cl_df, highlight_id=None, mode="pca"): x_col, y_col = (("tsne_x", "tsne_y") if mode == "tsne" else ("pca_x", "pca_y")) title = "t-SNE player style map" if mode == "tsne" else "PCA player style map" plot_df = cl_df.copy() plot_df["cluster"] = plot_df["cluster"].astype(int) plot_df["archetype"] = plot_df["cluster"].apply( lambda cid: CLUSTER_ARCHETYPES.get(cid, {}).get("name", f"Cluster {cid}") ) plot_df["hover_team"] = plot_df.get("team_name", "Unknown") plot_df["hover_comp"] = plot_df.get("competition_name", "Unknown") plot_df["hover_pos"] = plot_df.get("primary_position", "Unknown") fig = go.Figure() for cid in sorted(plot_df["cluster"].unique()): subset = plot_df[plot_df["cluster"] == cid] arch = CLUSTER_ARCHETYPES.get(int(cid), {}) name = arch.get("name", f"Cluster {cid}") color = arch.get("color", C["accent"]) fig.add_trace(go.Scattergl( x=subset[x_col], y=subset[y_col], mode="markers", name=name, marker=dict( size=8, color=color, opacity=0.68, line=dict(width=0.5, color="rgba(232,241,255,0.25)"), ), customdata=np.stack([ subset["display_name"].astype(str), subset["hover_pos"].astype(str), subset["hover_team"].astype(str), subset["hover_comp"].astype(str), ], axis=-1), hovertemplate=( "%{customdata[0]}
" "Type: " + name + "
" "Position: %{customdata[1]}
" "Team: %{customdata[2]}
" "Competition: %{customdata[3]}" ), )) if highlight_id is not None: highlighted = plot_df[plot_df["player_id"] == highlight_id] if not highlighted.empty: row = highlighted.iloc[0] fig.add_trace(go.Scattergl( x=[row[x_col]], y=[row[y_col]], mode="markers+text", name="Highlighted player", text=[row["display_name"]], textposition="top center", textfont=dict(family="Barlow Condensed", size=13, color=C["gold"]), marker=dict( size=18, color=C["gold"], symbol="star", line=dict(width=2, color=C["text"]), ), hovertemplate=( f"{row['display_name']}
" f"Type: {row['archetype']}
" f"Position: {row['hover_pos']}
" f"Team: {row['hover_team']}" ), )) fig.update_layout( **_PLOT_BASE(), title=dict( text=title, font=dict(family="Barlow Condensed", size=16, color=C["text"]), ), xaxis=dict( title="Style dimension 1", showgrid=True, gridcolor=C["border"], zeroline=False, tickfont=dict(color=C["muted"]), title_font=dict(color=C["muted"]), ), yaxis=dict( title="Style dimension 2", showgrid=True, gridcolor=C["border"], zeroline=False, tickfont=dict(color=C["muted"]), title_font=dict(color=C["muted"]), ), legend=dict( orientation="h", yanchor="bottom", y=1.02, xanchor="left", x=0, font=dict(family="Barlow Condensed", size=11, color=C["text"]), bgcolor="rgba(0,0,0,0)", ), height=560, margin=dict(t=92, b=40, l=20, r=20), ) return fig def _comparison_bars(r1, r2, n1, n2, pp_df): METRICS = [ ("Goals / 90", "goals_per90"), ("Assists / 90", "assists_per90"), ("xG / 90", "xg_per90"), ("xA / 90", "xassists_per90"), ("Prog. Passes / 90","progressive_passes_per90"), ("Dribbles / 90", "dribbles_completed_per90"), ("Tackles / 90", "tackles_per90"), ("Interceptions / 90","interceptions_per90"), ("Pass Accuracy", "pass_accuracy"), ("Pressures / 90", "pressures_per90"), ] labels, v1, v2 = [], [], [] for lbl, col in METRICS: if col in pp_df.columns: labels.append(lbl) v1.append( pct_rank(pp_df[col], r1.get(col, 0)) ) v2.append( pct_rank(pp_df[col], r2.get(col, 0)) ) st.markdown("""
Reading this chart — bars show percentile rank (0–100). Longer bar = better than more players at that stat. 50 = average.
""", unsafe_allow_html=True) fig = go.Figure() fig.add_trace(go.Bar( name=n1, x=labels, y=v1, marker_color=C["green"], text=[f"{v:.0f}" for v in v1], textposition="outside", textfont=dict( size=10, color=C["text"] ), hovertemplate="%{x}
" + n1 + ": %{y:.0f}th percentile" )) fig.add_trace(go.Bar( name=n2, x=labels, y=v2, marker_color=C["accent"], text=[f"{v:.0f}" for v in v2], textposition="outside", textfont=dict( size=10, color=C["text"] ), hovertemplate="%{x}
" + n2 + ": %{y:.0f}th percentile" )) fig.add_hline( y=50, line=dict( color=C["muted"], dash="dash", width=1 ), annotation_text="Average", annotation_position="right", annotation_font=dict( color=C["muted"], size=10 ) ) fig.update_layout( **_PLOT_BASE(), barmode="group", xaxis=dict( tickfont=dict( family="Barlow Condensed", size=10, color=C["text"] ) ), yaxis=dict( range=[0,125], gridcolor=C["border"], title=dict( text="Percentile Rank", font=dict( color=C["muted"] ) ) ), legend=dict( font=dict( family="Barlow Condensed", color=C["text"], size=13 ), bgcolor="rgba(0,0,0,0)" ), height=380, margin=dict(t=20,b=20,l=50,r=20) ) st.plotly_chart( fig, width="stretch", config={"displayModeBar": False} ) # ───────────────────────────────────────────────────────────────────────────── # PAGE 1 — SEARCH A PLAYER # ───────────────────────────────────────────────────────────────────────────── def page_search(ps_df, pp_df, cl_df): st.markdown('

🔍 Search a Player

', unsafe_allow_html=True) st.markdown('

Pick any player from the database and explore their full performance report

', unsafe_allow_html=True) st.markdown("
", unsafe_allow_html=True) # Guided picker player_row, per90_row, pid = player_picker(ps_df, pp_df, key_prefix="s") if player_row is None: return st.markdown("
", unsafe_allow_html=True) # ── Profile header ──────────────────────────────────────────────────────── render_profile_header(player_row, cl_df) # ── Key numbers strip ───────────────────────────────────────────────────── st.markdown("""
KEY NUMBERS AT A GLANCE
""", unsafe_allow_html=True) m1,m2,m3,m4,m5,m6 = st.columns(6) m1.metric("⚽ Goals", fmt(player_row.get("goals",0), 0), help="Total goals scored") m2.metric("🎯 Assists", fmt(player_row.get("assists",0), 0), help="Total goal assists") m3.metric("📈 xG", fmt(player_row.get("xg",0)), help="Expected Goals — how many goals the shots were worth") m4.metric("✨ xA", fmt(player_row.get("xassists",0)), help="Expected Assists — quality of chances created") m5.metric("⏱️ Minutes", fmt(player_row.get("minutes_played",0), 0), help="Total minutes played") m6.metric("🏁 Matches", fmt(player_row.get("matches_played",0), 0), help="Total matches appeared in") # Glossary always available st.markdown("
", unsafe_allow_html=True) glossary_expander() st.markdown("
", unsafe_allow_html=True) # ── Tabs ────────────────────────────────────────────────────────────────── t1, t2, t3 = st.tabs([ "📊 How do they rank?", "📋 All Statistics", "🎭 Playing Style", ]) with t1: st.markdown("""
What am I looking at?
Each bar shows how this player ranks compared to every other player in our database. A bar at 90th percentile means they are better than 90% of players at that stat. 50th = perfectly average. The dashed line marks the average.
""", unsafe_allow_html=True) c1, c2 = st.columns([3, 2]) with c1: st.plotly_chart(make_pct_chart(per90_row, pp_df), width="stretch", config={"displayModeBar": False}) with c2: # Strengths & weaknesses _render_strengths_weaknesses(per90_row, pp_df) with t2: _render_full_stats_beginner(player_row, per90_row) with t3: _render_playing_style(player_row, per90_row, pp_df, cl_df) def _render_strengths_weaknesses(per90_row, pp_df): CHECKS = { "Goals/90": "goals_per90", "Assists/90": "assists_per90", "xG/90": "xg_per90", "Prog. Passes/90": "progressive_passes_per90", "Dribbles/90": "dribbles_completed_per90", "Tackles/90": "tackles_per90", "Interceptions/90": "interceptions_per90", "Pressures/90": "pressures_per90", "Pass Accuracy": "pass_accuracy", "Aerial Duels/90": "aerial_duels_per90", } ranked = sorted( [(l, pct_rank(pp_df[c], per90_row.get(c,0))) for l,c in CHECKS.items() if c in pp_df.columns and c in per90_row.index], key=lambda x: x[1], reverse=True ) strong = ranked[:3] weak = ranked[-3:][::-1] st.markdown(f"""
💪 BIGGEST STRENGTHS
{''.join(f'''
✅ {l} {p:.0f}th percentile
''' for l,p in strong)}
⚠️ RELATIVE WEAKNESSES
These are still compared to ALL players — even a "weakness" can be fine for their position.
{''.join(f'''
△ {l} {p:.0f}th percentile
''' for l,p in weak)}
""", unsafe_allow_html=True) def _render_full_stats_beginner(player_row, per90_row): st.markdown("""
Reading this table
The left column shows career totals. The right shows per-90-minute rates — this is more useful for comparing players who played different amounts of time.
""", unsafe_allow_html=True) c1,c2,c3 = st.columns(3) with c1: st.markdown(f"""
⚽ ATTACKING
{stat_row("Goals (total)", fmt(player_row.get("goals",0),0), C["green"])} {stat_row("Shots (total)", fmt(player_row.get("shots",0),0))} {stat_row("Shots on Target", fmt(player_row.get("shots_on_target",0),0))} {stat_row("Shot Accuracy %", fmt(player_row.get("shot_accuracy",0)) + "%")} {stat_row("xG — total", fmt(player_row.get("xg",0)), C["gold"])}
Per 90 minutes:
{stat_row("Goals / 90 min", fmt(per90_row.get("goals_per90",0)), C["green"])} {stat_row("Shots / 90 min", fmt(per90_row.get("shots_per90",0)))} {stat_row("xG / 90 min", fmt(per90_row.get("xg_per90",0)), C["gold"])}
""", unsafe_allow_html=True) with c2: st.markdown(f"""
🎯 CREATING & PASSING
{stat_row("Assists (total)", fmt(player_row.get("assists",0),0), C["green"])} {stat_row("xA — total", fmt(player_row.get("xassists",0)), C["gold"])} {stat_row("Passes Attempted", fmt(player_row.get("passes_attempted",0),0))} {stat_row("Passes Completed", fmt(player_row.get("passes_completed",0),0))} {stat_row("Pass Accuracy %", fmt(player_row.get("pass_accuracy",0)) + "%", C["accent"])} {stat_row("Progressive Passes", fmt(player_row.get("progressive_passes",0),0))} {stat_row("Dribbles Won/Tried", f"{fmt(player_row.get('dribbles_completed',0),0)} / {fmt(player_row.get('dribbles_attempted',0),0)}")} {stat_row("Dribble Success %", fmt(player_row.get("dribble_success_rate",0)) + "%")}
Per 90 minutes:
{stat_row("Assists / 90", fmt(per90_row.get("assists_per90",0)), C["green"])} {stat_row("Prog. Passes / 90", fmt(per90_row.get("progressive_passes_per90",0)))}
""", unsafe_allow_html=True) with c3: st.markdown(f"""
🛡️ DEFENDING
{stat_row("Tackles (total)", fmt(player_row.get("tackles",0),0))} {stat_row("Tackles Won", fmt(player_row.get("tackles_won",0),0))} {stat_row("Tackle Success %", fmt(player_row.get("tackle_success_rate",0)) + "%", C["accent"])} {stat_row("Interceptions", fmt(player_row.get("interceptions",0),0), C["accent"])} {stat_row("Pressures", fmt(player_row.get("pressures",0),0))} {stat_row("Clearances", fmt(player_row.get("clearances",0),0))} {stat_row("Aerial Duels Won/Total", f"{fmt(player_row.get('aerial_duels_won',0),0)} / {fmt(player_row.get('aerial_duels',0),0)}")} {stat_row("Aerial Success %", fmt(player_row.get("aerial_success_rate",0)) + "%")}
Per 90 minutes:
{stat_row("Tackles / 90", fmt(per90_row.get("tackles_per90",0)))} {stat_row("Interceptions / 90", fmt(per90_row.get("interceptions_per90",0)))}
""", unsafe_allow_html=True) def _render_playing_style(player_row, per90_row, pp_df, cl_df): arch = {} if cl_df is not None: cl_rows = cl_df[cl_df["player_id"] == player_row["player_id"]] if not cl_rows.empty: try: arch = CLUSTER_ARCHETYPES.get(int(cl_rows.iloc[0]["cluster"]), {}) except Exception: pass ac = arch.get("color", C["accent"]) an = arch.get("name", "Unknown") ai = arch.get("icon", "⚽") ad = arch.get("desc", "") def idx(cols): scores = [pct_rank(pp_df[c], per90_row.get(c,0)) for c in cols if c in pp_df.columns and c in per90_row.index] return np.mean(scores) if scores else 50.0 att = idx(["goals_per90","shots_per90","xg_per90","shot_accuracy"]) cre = idx(["assists_per90","progressive_passes_per90","dribbles_completed_per90","pass_accuracy"]) dfe = idx(["tackles_per90","interceptions_per90","pressures_per90","clearances_per90"]) def index_bar(label, val, desc): col = pct_color(val) badge = ("Elite" if val>=85 else "Strong" if val>=65 else "Average" if val>=40 else "Low") return f"""
{label} {val:.0f} / 100 — {badge}
{desc}
""" st.markdown(f"""
{ai}
Player Type: {an}
{ad}
Determined by machine learning — comparing their statistical fingerprint to 7,000+ players.
PLAYING STYLE INDICES — scored out of 100 vs all players
{index_bar("⚽ Attacking Threat", att, "Goals, shots, and xG quality")} {index_bar("🎨 Creative Influence", cre, "Assists, progressive passes, dribbles, passing accuracy")} {index_bar("🛡️ Defensive Work", dfe, "Tackles, interceptions, pressures, clearances")}
""", unsafe_allow_html=True) # ───────────────────────────────────────────────────────────────────────────── # PAGE 2 — COMPARE TWO PLAYERS # ───────────────────────────────────────────────────────────────────────────── def page_compare(ps_df, pp_df, cl_df): st.markdown('

⚖️ Compare Two Players

', unsafe_allow_html=True) st.markdown('

Pick any two players and see how they stack up side-by-side

', unsafe_allow_html=True) st.markdown("
", unsafe_allow_html=True) # Explanation banner st.markdown("""
How comparison works
Use the filters below to find Player 1, then scroll down to find Player 2. The radar chart shows each player's percentile rank (0–100) — so a bigger shape means a stronger all-round player. The bar chart lets you compare individual stats head-to-head.
""", unsafe_allow_html=True) # Player 1 picker st.markdown("""
👤 PLAYER 1 (Green)
""", unsafe_allow_html=True) p1_row, p1_pp, p1_id = player_picker(ps_df, pp_df, key_prefix="cmp1", show_howto=False) st.markdown("
", unsafe_allow_html=True) # Player 2 picker st.markdown("""
👤 PLAYER 2 (Blue)
""", unsafe_allow_html=True) p2_row, p2_pp, p2_id = player_picker(ps_df, pp_df, key_prefix="cmp2", show_howto=False) if p1_row is None or p2_row is None: return n1 = p1_row.get("display_name", "Player 1") n2 = p2_row.get("display_name", "Player 2") if p1_id == p2_id: st.warning("⚠️ You selected the same player twice. Please choose two different players.") return st.markdown("
", unsafe_allow_html=True) # Side-by-side header cards col1, col2 = st.columns(2) with col1: _mini_card(p1_row, cl_df, C["green"]) with col2: _mini_card(p2_row, cl_df, C["accent"]) st.markdown("
", unsafe_allow_html=True) glossary_expander() st.markdown("
", unsafe_allow_html=True) t1, t2, t3 = st.tabs(["🕸️ Radar Chart", "📊 Side-by-Side Stats", "📋 Full Table"]) with t1: st.markdown("""
Reading the radar — Each point on the web = a stat category. Further from centre = higher percentile. A bigger overall shape = stronger player in that area. Numbers shown are percentile ranks (0–100), not raw stat values.
""", unsafe_allow_html=True) st.plotly_chart(make_radar(p1_pp, p2_pp, n1, n2, pp_df), width="stretch", config={"displayModeBar": False}) with t2: _comparison_bars(p1_pp, p2_pp, n1, n2, pp_df) with t3: _comparison_table(p1_row, p2_row, p1_pp, p2_pp, n1, n2, pp_df) def _mini_card(ps_row, cl_df, color): arch = {} if cl_df is not None: cl_rows = cl_df[cl_df["player_id"] == ps_row["player_id"]] if not cl_rows.empty: try: arch = CLUSTER_ARCHETYPES.get(int(cl_rows.iloc[0]["cluster"]), {}) except Exception: pass ac = arch.get("color", color) an = arch.get("name", "—") ai = arch.get("icon", "⚽") name = ps_row.get("display_name", ps_row.get("player_name","?")) st.markdown(f"""
{name}
{ps_row.get('team','—')} · {ps_row.get('position','—')} · {ps_row.get('competition','—')}
⚽ Goals: {fmt(ps_row.get('goals',0),0)}
🎯 Assists: {fmt(ps_row.get('assists',0),0)}
📈 xG: {fmt(ps_row.get('xg',0))}
⏱️ Mins: {fmt(ps_row.get('minutes_played',0),0)}
{ai} {an}
""", unsafe_allow_html=True) def _comparison_table(ps1, ps2, pp1, pp2, n1, n2, pp_df): st.markdown("""
Reading this table — Green = that player wins this stat. Numbers in brackets e.g. (82nd) = percentile rank.
""", unsafe_allow_html=True) rows = [] STATS = [ ("Goals (total)", "goals", ps1,ps2, False), ("Assists (total)", "assists", ps1,ps2, False), ("xG (total)", "xg", ps1,ps2, False), ("xA (total)", "xassists", ps1,ps2, False), ("Pass Accuracy %", "pass_accuracy", ps1,ps2, False), ("Goals / 90", "goals_per90", pp1,pp2, True), ("Assists / 90", "assists_per90", pp1,pp2, True), ("xG / 90", "xg_per90", pp1,pp2, True), ("Prog. Passes / 90","progressive_passes_per90", pp1,pp2, True), ("Dribbles / 90", "dribbles_completed_per90", pp1,pp2, True), ("Tackles / 90", "tackles_per90", pp1,pp2, True), ("Interceptions / 90","interceptions_per90", pp1,pp2, True), ("Pressures / 90", "pressures_per90", pp1,pp2, True), ] for label, col, src1, src2, is_per90 in STATS: v1 = src1.get(col, np.nan); v2 = src2.get(col, np.nan) if pd.isna(v1) and pd.isna(v2): continue pct_col = col if is_per90 else col p1 = pct_rank(pp_df[pct_col], v1) if pct_col in pp_df.columns and not pd.isna(v1) else None p2 = pct_rank(pp_df[pct_col], v2) if pct_col in pp_df.columns and not pd.isna(v2) else None rows.append({ "Stat": label, n1: f"{fmt(v1)} {'('+str(int(p1))+'th)' if p1 is not None else ''}", n2: f"{fmt(v2)} {'('+str(int(p2))+'th)' if p2 is not None else ''}", "_v1": float(v1) if not pd.isna(v1) else -999, "_v2": float(v2) if not pd.isna(v2) else -999, }) df = pd.DataFrame(rows) display = df[["Stat", n1, n2]].copy() def highlight(row): original = df[df["Stat"] == row["Stat"]] if original.empty: return ["","",""] v1 = original["_v1"].values[0]; v2 = original["_v2"].values[0] if v1 > v2: return ["","background-color:rgba(0,255,157,0.12);color:#00ff9d;", "background-color:rgba(255,107,53,0.06);color:#8a9ab0;"] elif v2 > v1: return ["","background-color:rgba(255,107,53,0.06);color:#8a9ab0;", "background-color:rgba(0,212,255,0.12);color:#00d4ff;"] return ["","",""] st.dataframe(display.style.apply(highlight, axis=1), width="stretch", height=480) # ───────────────────────────────────────────────────────────────────────────── # PAGE 3 — EXPLORE PLAYER TYPES # ───────────────────────────────────────────────────────────────────────────── def page_clusters(ps_df, cl_df): st.markdown('

🗺️ Explore Player Types

', unsafe_allow_html=True) st.markdown('

Our AI grouped 7,000+ players into 8 playing styles using machine learning — explore them here

', unsafe_allow_html=True) st.markdown("
", unsafe_allow_html=True) # Explanation st.markdown("""
What is this?
We analysed 24 statistics for every player and used a machine-learning technique called KMeans clustering to automatically group players by how they play. Players in the same group have similar statistical fingerprints — not the same team or nationality, just a similar playing style.

The chart below shows every player as a dot. Dots close together = players who play similarly. Hover over any dot to see the player name.
""", unsafe_allow_html=True) # Controls row c1, c2, c3 = st.columns([3,2,2]) with c1: st.markdown('Highlight a specific player (optional)', unsafe_allow_html=True) eligible_ids = set(ps_df.loc[ps_df["minutes_played"] >= MIN_MINUTES, "player_id"]) highlight_df = cl_df[cl_df["player_id"].isin(eligible_ids)].copy() all_names = ["— Don't highlight anyone —"] + sorted(highlight_df["display_name"].dropna().unique()) hl_name = st.selectbox("hl", all_names, label_visibility="collapsed", help="Search for a player to mark them with a gold star on the chart") with c2: st.markdown('Filter to one player type', unsafe_allow_html=True) type_opts = ["Show all types"] + [f"{a['icon']} {a['name']}" for a in CLUSTER_ARCHETYPES.values()] type_filter = st.selectbox("tf", type_opts, label_visibility="collapsed") with c3: st.markdown('Chart style', unsafe_allow_html=True) proj = st.radio("proj", ["PCA (faster)", "t-SNE (richer)"], horizontal=True, label_visibility="collapsed") # Resolve highlight hl_id = None if hl_name != "— Don't highlight anyone —": rows = highlight_df[highlight_df["display_name"] == hl_name] if not rows.empty: hl_id = int(rows.iloc[0]["player_id"]) # Filter by type plot_df = cl_df.copy() if type_filter != "Show all types": target_name = type_filter.split(" ", 1)[1] # strip emoji cid = next((k for k,v in CLUSTER_ARCHETYPES.items() if v["name"] == target_name), None) if cid is not None: plot_df = plot_df[plot_df["cluster"] == cid] mode = "tsne" if "t-SNE" in proj else "pca" st.plotly_chart(make_cluster_scatter(plot_df, hl_id, mode), width="stretch", config={"displayModeBar": False}) st.markdown('
', unsafe_allow_html=True) # Archetype cards — in a 4-column grid st.markdown("""
THE 8 PLAYER TYPES EXPLAINED
""", unsafe_allow_html=True) for row_items in [list(CLUSTER_ARCHETYPES.items())[i:i+4] for i in range(0,8,4)]: cols = st.columns(4) for col, (cid, arch) in zip(cols, row_items): count = len(cl_df[cl_df["cluster"] == cid]) with col: st.markdown(f"""
{arch['icon']}
{arch['name']}
{count:,} players in this group
{arch['desc']}
""", unsafe_allow_html=True) # Top 10 players in selected type if type_filter != "Show all types": target_name = type_filter.split(" ", 1)[1] cid = next((k for k,v in CLUSTER_ARCHETYPES.items() if v["name"] == target_name), None) if cid is not None: st.markdown(f"""
Top Players — {type_filter}
""", unsafe_allow_html=True) pids = cl_df[cl_df["cluster"]==cid]["player_id"].tolist() top = (ps_df[ps_df["player_id"].isin(pids)] .sort_values("minutes_played", ascending=False) .drop_duplicates("display_name") .head(15)[["display_name","team","competition","position", "goals","assists","minutes_played"]]) top.columns = ["Player","Team","Competition","Position","Goals","Assists","Minutes"] st.dataframe(top.reset_index(drop=True), width="stretch", height=420) # ───────────────────────────────────────────────────────────────────────────── # PAGE 4 — FIND SIMILAR PLAYERS # ───────────────────────────────────────────────────────────────────────────── def page_similar(ps_df, pp_df, sim_df, cl_df): st.markdown('

🔗 Find Similar Players

', unsafe_allow_html=True) st.markdown('

Pick any player and we\'ll find the players who play most like them — great for scouting alternatives

', unsafe_allow_html=True) st.markdown("
", unsafe_allow_html=True) st.markdown("""
How this works
We compare the player's statistical profile across 24 metrics against every other player using a technique called cosine similarity. A score of 100% = statistically identical. A score of 80%+ = very similar playing style. This doesn't mean they're equally talented — just that they play in a similar way.
""", unsafe_allow_html=True) # Reference player picker player_row, per90_row, pid = player_picker(ps_df, pp_df, key_prefix="sim", show_howto=False) if player_row is None: return c1,c2 = st.columns([2,1]) with c1: n_results = st.slider("How many similar players to show?", 5, 20, 10, help="Slide to see more or fewer results") with c2: same_pos = st.toggle("Same position only", help="Turn this ON to only see players in the same position") if st.button("🔍 Find Similar Players", width="content"): with st.spinner("Comparing against 7,000+ players…"): sim_rows = sim_df[sim_df["player_id"] == pid] if sim_rows.empty: st.error("Player not found in similarity dataset.") return X = sim_df[FEATURE_COLS_SIM].values idx_val = sim_rows.index[0] sims = cosine_similarity(X[idx_val].reshape(1,-1), X)[0] result = sim_df.copy(); result["similarity"] = sims result = result[result["player_id"] != pid] if same_pos: my_pos = sim_rows.iloc[0]["primary_position"] result = result[result["primary_position"] == my_pos] eligible_ids = set(ps_df.loc[ps_df["minutes_played"] >= MIN_MINUTES, "player_id"]) result = result[result["player_id"].isin(eligible_ids)] result = result.nlargest(n_results, "similarity").reset_index(drop=True) st.markdown("
", unsafe_allow_html=True) # Reference card render_profile_header(player_row, cl_df) st.markdown(f"""
Showing the {len(result)} most statistically similar players {'in the same position' if same_pos else 'from all positions'}
""", unsafe_allow_html=True) # Result cards in 2-column grid for i in range(0, len(result), 2): cols = st.columns(2) for j, col in enumerate(cols): idx2 = i+j if idx2 >= len(result): break row = result.iloc[idx2] score = int(row["similarity"] * 100) scol = (C["green"] if score>=90 else C["accent"] if score>=75 else C["gold"]) # Archetype cl_rows = cl_df[cl_df["player_id"] == row["player_id"]] if cl_df is not None else pd.DataFrame() arch = {} if not cl_rows.empty: try: arch = CLUSTER_ARCHETYPES.get(int(cl_rows.iloc[0]["cluster"]), {}) except Exception: pass # Plain English similarity label sim_label = ("🟢 Virtually identical" if score>=92 else "✅ Very similar" if score>=80 else "🔵 Somewhat similar" if score>=65 else "⚪ Loosely similar") with col: st.markdown(f"""
#{idx2+1}   {row['display_name']}
{row['primary_position']} · {row['team_name']}
{row['competition_name']}
{score}%
SIMILARITY
{sim_label}
{arch.get('icon','⚽')} {arch.get('name','—')}
""", unsafe_allow_html=True) else: st.markdown(f"""
👆 Select a player above, then click Find Similar Players
""", unsafe_allow_html=True) # ───────────────────────────────────────────────────────────────────────────── # MAIN # ───────────────────────────────────────────────────────────────────────────── def main(): with st.spinner("Loading player data…"): ps_df = load_player_stats() pp_df = load_per90() sim_df = load_similarity() cl_df = load_clusters() page = render_sidebar() if "Search" in page: page_search(ps_df, pp_df, cl_df) elif "Compare" in page: page_compare(ps_df, pp_df, cl_df) elif "Explore" in page or "Types" in page: page_clusters(ps_df, cl_df) elif "Similar" in page: page_similar(ps_df, pp_df, sim_df, cl_df) if __name__ == "__main__": main()