NovaCorp People Analytics Report

# Source: generate_viz.py  -  chart1_retention_curve()
df = pd.read_csv(DATA_DIR / 'employees.csv')
df['event'] = (df['status'] == 'departed').astype(int)
df['time'] = df['tenure_months'].clip(upper=120)

from lifelines import KaplanMeierFitter
fig, ax = plt.subplots(figsize=(10, 6))
depts = df['department'].value_counts().head(7).index.tolist()
for i, dept in enumerate(depts):
    sub = df[df['department'] == dept]
    kmf = KaplanMeierFitter()
    kmf.fit(sub['time'], sub['event'], label=dept)
    kmf.plot_survival_function(ax=ax, color=COLORS_10[i], linewidth=2)

ax.set_title('NovaCorp 员工留存曲线（按部门）', fontsize=16, fontweight='bold')
ax.set_xlabel('在职月数'); ax.set_ylabel('留存概率')
ax.set_ylim(0.4, 1.02); ax.set_xlim(0, 120)
ax.legend(loc='lower left', fontsize=9, ncol=2)
fig.savefig(VIZ_DIR / '01_retention_curve.png')

# Source: generate_viz.py  -  chart8_push_pull_pathway()
df = pd.read_csv(OUT_01 / 'pathway_dept.csv')
df = df.sort_values('push_share', ascending=True)

fig, ax = plt.subplots(figsize=(10, 6))
bar_height = 0.5
ax.barh(y_pos, df['push_share'], bar_height, color=ACCENT, alpha=0.8, label='Push')
ax.barh(y_pos, 1 - df['push_share'], bar_height, left=df['push_share'],
        color=TEAL, alpha=0.8, label='Pull')

for i, (_, row) in enumerate(df.iterrows()):
    ax.text(row['push_share']/2, i, f'{row["push"]}人', ha='center', va='center', fontsize=9)
    ax.text(row['push_share']+(1-row['push_share'])/2, i, f'{row["pull"]}人',
            ha='center', va='center', fontsize=9)

ax.set_yticks(y_pos); ax.set_yticklabels(df['department'].values)
ax.set_title('NovaCorp 离职动因分析  -  Push vs Pull（按部门）')
ax.legend(loc='lower right'); ax.set_xlim(0, 1)
fig.savefig(VIZ_DIR / '08_push_pull_pathway.png')

# Source: generate_viz.py  -  chart9_regrettable_heatmap()
df = pd.read_csv(OUT_01 / 'regrettable_dept_entity.csv')
pivot = df.pivot_table(index='department', columns='legacy_entity_code',
                       values='regrettable_share', aggfunc='first')

fig, ax = plt.subplots(figsize=(9, 6))
im = ax.imshow(pivot.values, cmap='YlOrRd', aspect='auto')
for i in range(pivot.shape[0]):
    for j in range(pivot.shape[1]):
        val = pivot.iloc[i, j]
        if not np.isnan(val):
            ax.text(j, i, f'{val:.1%}', ha='center', va='center',
                    color='white' if val > 0.18 else '#333')

ax.set_xticks(range(pivot.shape[1])); ax.set_xticklabels(pivot.columns)
ax.set_yticks(range(pivot.shape[0])); ax.set_yticklabels(pivot.index)
ax.set_title('NovaCorp 遗憾离职率热力图  -  部门 × 实体')
fig.colorbar(im, ax=ax, shrink=0.8)
fig.savefig(VIZ_DIR / '09_regrettable_heatmap.png')

# Source: code/p0-1_descriptive.py  -  Fig1 Entity 离职率
e2 = ent.copy(); e2["rate"] = e2["rate"] * 100
fig, ax = plt.subplots(figsize=(8, 5))
colors = [RED if e == "Entity_B" else BLUE for e in e2["legacy_entity_code"]]
bar(ax, e2["legacy_entity_code"], e2["rate"],
    "各 Entity 自愿离职率 (voluntary only)", color=colors, rot=20)
ax.axhline(8.45, color=GRAY, ls="--", lw=1)
ax.text(0, 8.6, "全司 8.45%", fontsize=8, color=GRAY)
plt.tight_layout(); plt.savefig(f"{OUT_DIR}/fig1_entity_rate.png", dpi=120)

# Source: code/p0-1_descriptive.py  -  Fig2 Department×Entity 热力图
piv = de.pivot_table(index="department", columns="legacy_entity_code", values="rate")
piv = piv.reindex(piv.mean(axis=1).sort_values(ascending=False).index)
fig, ax = plt.subplots(figsize=(9, 7))
im = ax.imshow(piv.values, cmap="RdYlGn_r", aspect="auto")
ax.set_xticks(range(piv.shape[1])); ax.set_xticklabels(piv.columns, fontsize=9)
ax.set_yticks(range(piv.shape[0])); ax.set_yticklabels(piv.index, fontsize=9)
for i in range(piv.shape[0]):
    for j in range(piv.shape[1]):
        v = piv.values[i, j] * 100
        ax.text(j, i, f"{v:.0f}" if not np.isnan(v) else "-",
                ha="center", va="center", fontsize=8,
                color="white" if v > 12 else "black")
ax.set_title("Department × Entity 自愿离职率 (%)", fontweight="bold")
plt.colorbar(im, ax=ax, label="离职率 (%)")
plt.savefig(f"{OUT_DIR}/fig2_dept_entity_heatmap.png", dpi=120)

# Source: code/p0-1_descriptive.py  -  Fig3 role_level × role_family
piv3 = rl.pivot_table(index="role_level", columns="role_family", values="rate")
fig, ax = plt.subplots(figsize=(10, 5.5))
im = ax.imshow(piv3.values, cmap="RdYlGn_r", aspect="auto")
ax.set_xticks(range(piv3.shape[1]))
ax.set_xticklabels(piv3.columns, fontsize=8, rotation=30, ha="right")
ax.set_yticks(range(piv3.shape[0]))
ax.set_yticklabels(piv3.index, fontsize=9)
for i in range(piv3.shape[0]):
    for j in range(piv3.shape[1]):
        v = piv3.values[i, j] * 100
        ax.text(j, i, f"{v:.0f}" if not np.isnan(v) else "-",
                ha="center", va="center", fontsize=7.5,
                color="white" if (not np.isnan(v) and v > 12) else "black")
ax.set_title("Role Level × Role Family 自愿离职率 (%)", fontweight="bold")
plt.colorbar(im, ax=ax, label="离职率 (%)")
plt.savefig(f"{OUT_DIR}/fig3_role_level_family.png", dpi=120)

# Source: code/p0-1_descriptive.py  -  Fig4 age_band × gender
ag32 = ag.pivot_table(index="age_band", columns="gender", values="rate")
gend_order = ["Male", "Female", "Non-binary", "Prefer not to say"]
ag32 = ag32[[c for c in gend_order if c in ag32.columns]].loc[age_order]
fig, ax = plt.subplots(figsize=(10, 5.5))
x = np.arange(len(ag32.index)); w = 0.8 / len(ag32.columns)
colmap = {"Male": "#3498db", "Female": "#e74c3c",
          "Non-binary": "#2ecc71", "Prefer not to say": GRAY}
for k, g in enumerate(ag32.columns):
    vals = ag32[g].fillna(0)
    ax.bar(x + (k-len(ag32.columns)/2+0.5)*w, vals, w,
           label=g, color=colmap.get(g, GRAY))
    for xi, yi in zip(x, vals):
        if yi > 0:
            ax.text(xi+(k-len(ag32.columns)/2+0.5)*w, yi,
                    f"{yi:.0f}", ha="center", va="bottom", fontsize=7)
ax.set_xticks(x); ax.set_xticklabels(ag32.index)
ax.set_ylabel("离职率 (%)"); ax.set_title("Age Band × Gender 自愿离职率 (%)")
ax.legend(); plt.savefig(f"{OUT_DIR}/fig4_age_gender.png", dpi=120)

# Source: code/p0-1_descriptive.py  -  Fig5 contract_type
c5 = ct.copy(); c5["rate"] = c5["rate"] * 100
fig, ax = plt.subplots(figsize=(7, 4.5))
bar(ax, c5["contract_type"], c5["rate"],
    "Contract Type 自愿离职率 (%)", rot=15)
plt.tight_layout()
plt.savefig(f"{OUT_DIR}/fig5_contract.png", dpi=120)

# Source: code/p0-1_descriptive.py  -  Fig6 tenure bucket
def tenure_bucket(t):
    if t < 12:   return "<12月"
    if t < 24:   return "12-24月"
    if t < 36:   return "24-36月"
    if t < 60:   return "36-60月"
    return "60+月"
base["tenure_bucket"] = base["tenure_months"].map(tenure_bucket)

t6 = tb.copy(); t6["rate"] = t6["rate"] * 100
fig, ax = plt.subplots(figsize=(8, 5))
bar(ax, t6["tenure_bucket"].astype(str), t6["rate"],
    "Tenure 分段自愿离职率 (%)", color=BLUE, rot=15)
plt.savefig(f"{OUT_DIR}/fig6_tenure.png", dpi=120)

# Source: code/p0-1_descriptive.py  -  Fig7 薪资分布对比
fig, ax = plt.subplots(figsize=(9, 5))
bins = np.linspace(0, max(active_sal.max(), exit_sal.max()), 60)
ax.hist(active_sal/1000, bins=bins/1000, alpha=0.6,
        label=f"在职 (n={len(active_sal):,})", color=BLUE)
ax.hist(exit_sal/1000, bins=bins/1000, alpha=0.6,
        label=f"自愿离职者-离职薪资 (n={len(exit_sal):,})", color=RED)
ax.set_xlabel("薪资 (千)"); ax.set_ylabel("人数")
ax.set_title("自愿离职者 vs 在职者 薪资分布", fontweight="bold")
ax.legend(); ax.grid(alpha=0.3)
plt.savefig(f"{OUT_DIR}/fig7_salary_dist.png", dpi=120)

# Source: code/p0-1_descriptive.py  -  Fig8 regrettable by role_family
rf = reg_by_fam.head(7)
fig, ax = plt.subplots(figsize=(8, 5))
bar(ax, rf["role_family"], rf["regrettable_share"] * 100,
    "自愿离职中 Regrettable 占比 by Role Family", color=RED, rot=25)
plt.tight_layout()
plt.savefig(f"{OUT_DIR}/fig8_regrettable_family.png", dpi=120)

# Source: code/p0-1_descriptive.py  -  Fig9 pathway push share by dept
pd9 = pw_dept.sort_values("push_share", ascending=False).head(12)
fig, ax = plt.subplots(figsize=(9, 5))
bar(ax, pd9["department"], pd9["push_share"] * 100,
    "Push(可预防)型占比 Top 部门", color="#e67e22", rot=30)
for xi, yi in zip(range(len(pd9)), pd9["push_share"] * 100):
    ax.text(xi, yi, f"{yi:.0f}%", ha="center", va="bottom", fontsize=8)
plt.savefig(f"{OUT_DIR}/fig9_pathway_push.png", dpi=120)

# Source: generate_viz.py  -  chart3_odds_ratio_forest()
df = pd.read_csv(OUT_DIR / 'logreg_odds_ratio.csv')
df = df[df['or_low'] > 0]  # filter extreme values
df = df[(df['or_high'] < 100) & (df['odds_ratio'] < 50)]
df = df.sort_values('odds_ratio', ascending=True)

fig, ax = plt.subplots(figsize=(12, 8))
y_pos = range(len(df_plot))
for i, (_, row) in enumerate(df_plot.iterrows()):
    or_val = row['odds_ratio']; low = row['or_low']; high = row['or_high']
    color = ACCENT if or_val > 1 else TEAL
    ax.errorbar(or_val, i, xerr=[[or_val - low], [high - or_val]],
                fmt='o', color=color, capsize=3, capthick=1.5, markersize=7)
ax.set_xscale('log'); ax.set_xlim(0.3, 10)
ax.axvline(x=1, color='white', linestyle='--', alpha=0.6)
ax.set_title('NovaCorp 自愿离职逻辑回归  -  OR森林图')
fig.savefig(VIZ_DIR / '03_odds_ratio_forest.png')

# Source: code/p0-2_modeling.py  -  SHAP analysis
import shap
explainer = shap.TreeExplainer(xgb)
shap_values = explainer.shap_values(X_te_p)
shap_df = pd.DataFrame({
    "feature": feat_names,
    "mean_abs_shap": np.abs(shap_values).mean(axis=0),
}).sort_values("mean_abs_shap", ascending=False)

# SHAP summary plot  -  每个样本每个特征的贡献
fig, ax = plt.subplots(figsize=(10, 9))
shap.summary_plot(shap_values, X_te_p, feature_names=feat_names,
                  max_display=20, show=False)
plt.tight_layout()
plt.savefig(f"{OUT_DIR}/fig_shap_summary.png", dpi=120, bbox_inches="tight")

# Source: code/p0-2_modeling.py  -  SHAP bar plot
fig, ax = plt.subplots(figsize=(9, 7))
shap.summary_plot(shap_values, X_te_p, feature_names=feat_names,
                  plot_type="bar", max_display=20, show=False)
plt.tight_layout()
plt.savefig(f"{OUT_DIR}/fig_shap_bar.png", dpi=120, bbox_inches="tight")
plt.close()
print("SHAP 图已保存")

# Source: generate_viz.py  -  chart2_correlation_heatmap()
eng = pd.read_csv(DATA_DIR / 'engagement.csv')
perf = pd.read_csv(DATA_DIR / 'performance.csv')
eng_cols = ['manager_effectiveness','psychological_safety','recognition',
            'career_development','senior_leadership_trust','purpose_meaning',
            'wellbeing','confidence_in_role_future']

eng_agg = eng.groupby('employee_id')[eng_cols].mean().reset_index()
perf_agg = perf.groupby('employee_id')['goal_achievement_score'].mean().reset_index()
merged = eng_agg.merge(perf_agg, on='employee_id', how='inner')
all_cols = eng_cols + ['goal_achievement_score']
corr = merged[all_cols].corr()

labels_cn = ['管理者效能','心理安全','认可感','职业发展',
             '高管信任','目标意义','身心健康','岗位信心','目标达成']

fig, ax = plt.subplots(figsize=(11, 9))
im = ax.imshow(corr.values, cmap='RdYlBu_r', vmin=-1, vmax=1, aspect='auto')
for i in range(len(corr)):
    for j in range(len(corr)):
        ax.text(j, i, f'{corr.iloc[i, j]:.2f}', ha='center', va='center', fontsize=9,
                color='white' if abs(corr.iloc[i, j]) > 0.5 else '#ccc')
ax.set_xticks(range(len(labels_cn))); ax.set_yticks(range(len(labels_cn)))
ax.set_xticklabels(labels_cn, rotation=45, ha='right', fontsize=10)
ax.set_yticklabels(labels_cn, fontsize=10)
fig.colorbar(im, ax=ax, shrink=0.8).set_label('Pearson r')
fig.savefig(VIZ_DIR / '02_correlation_heatmap.png')

# Source: generate_viz.py  -  chart6_interaction_heatmap()
df = pd.read_csv(OUT_DIR / 'interact_mgr_recog.csv')
pivot = df.pivot_table(index='mgr_q', columns='recog_hi',
                       values='mean', aggfunc='first')
pivot_size = df.pivot_table(index='mgr_q', columns='recog_hi',
                            values='size', aggfunc='first')

fig, ax = plt.subplots(figsize=(8, 6))
im = ax.imshow(pivot.values, cmap='RdYlGn_r', aspect='auto')
for i in range(pivot.shape[0]):
    for j in range(pivot.shape[1]):
        val = pivot.iloc[i, j]; sz = pivot_size.iloc[i, j]
        ax.text(j, i, f'{val:.1%}\nn={int(sz)}', ha='center', va='center',
                fontsize=11, color='white' if val > 0.08 else '#ccc')

ax.set_xticks([0,1]); ax.set_xticklabels(['低认可','高认可'])
ax.set_yticks(range(len(pivot.index)))
ax.set_yticklabels([f'管理者效能 Q{i}' for i in pivot.index])
ax.set_title('NovaCorp 管理者效能 × 认可感交互效应\n（离职率）')
fig.colorbar(im, ax=ax, shrink=0.8)
fig.savefig(VIZ_DIR / '06_interaction_heatmap.png')

# Source: generate_viz.py  -  chart10_age_gender_attrition()
df = pd.read_csv(OUT_01 / 'age_gender_rate.csv')
age_order = ['18-24','25-29','30-34','35-39','40-44','45-49','50-54','55-59','60+']
gender_order = ['Female','Male','Non-binary','Prefer not to say']
gender_cn = {'Female':'女','Male':'男','Non-binary':'非二元','Prefer not to say':'不愿透露'}

pivot = df.pivot_table(index='age_band', columns='gender', values='rate', aggfunc='first')
pivot = pivot.reindex([a for a in age_order if a in pivot.index])
pivot = pivot[[g for g in gender_order if g in pivot.columns]]

fig, ax = plt.subplots(figsize=(12, 7))
x = np.arange(len(pivot.index)); width = 0.2
n_groups = len(pivot.columns)
offsets = np.linspace(-(n_groups-1)*width/2, (n_groups-1)*width/2, n_groups)

for i, col in enumerate(pivot.columns):
    ax.bar(x + offsets[i], pivot[col], width, label=gender_cn.get(col, col),
           color=COLORS_10[i], alpha=0.85, edgecolor='white', linewidth=0.5)

ax.set_xticks(x); ax.set_xticklabels(pivot.index)
ax.set_ylabel('自愿离职率 (%)'); ax.legend()
overall = df['rate'].mean()
ax.axhline(y=overall, color=GOLD, linestyle='--', linewidth=1.5, alpha=0.8,
           label=f'整体均值: {overall:.1f}%')
fig.savefig(VIZ_DIR / '10_age_gender_attrition.png')

# Source: generate_viz.py  -  chart5_engagement_trend()
eng = pd.read_csv(DATA_DIR / 'engagement.csv')
eng_cols = ['manager_effectiveness','psychological_safety','recognition',
            'career_development','senior_leadership_trust','purpose_meaning',
            'wellbeing','confidence_in_role_future']
trends = eng.groupby('wave_number')[eng_cols].mean()

labels_cn = ['管理者效能','心理安全','认可感','职业发展',
             '高管信任','目标意义','身心健康','岗位信心']

fig, ax = plt.subplots(figsize=(12, 7))
waves = trends.index
for i, col in enumerate(eng_cols):
    ax.plot(waves, trends[col], 'o-', color=COLORS_10[i], linewidth=2,
            markersize=7, label=labels_cn[i])

ax.set_xlabel('调查波次'); ax.set_ylabel('平均得分 (1-5)')
ax.set_title('NovaCorp 敬业度维度趋势 (2024-2025)')
ax.set_xticks(waves); ax.set_xticklabels([f'第{w}波' for w in waves])
ax.set_ylim(3.2, 4.8); ax.legend(loc='lower right', fontsize=9, ncol=2)
fig.savefig(VIZ_DIR / '05_engagement_trend.png')

# Source: generate_viz.py  -  chart4_salary_boxplot()
df = pd.read_csv(DATA_DIR / 'employees.csv')
df = df[df['salary'].between(30000, 500000)]
order = df.groupby('role_family')['salary'].median().sort_values(ascending=False).index.tolist()

fig, ax = plt.subplots(figsize=(12, 7))
bp = ax.boxplot([df[df['role_family']==rf]['salary'].dropna().values for rf in order],
                patch_artist=True, widths=0.6,
                medianprops={'color': 'white', 'linewidth': 2},
                whiskerprops={'color': '#aaa'},
                capprops={'color': '#aaa'},
                flierprops={'marker': '.', 'alpha': 0.3, 'markersize': 2})

for i, patch in enumerate(bp['boxes']):
    patch.set_facecolor(COLORS_10[i % len(COLORS_10)])
    patch.set_alpha(0.7)

ax.set_xticklabels(order, rotation=30, ha='right', fontsize=10)
ax.set_ylabel('年薪 (AUD)')
ax.set_title('NovaCorp 薪资分布  -  按职能族')
ax.yaxis.set_major_formatter(mticker.FuncFormatter(lambda x, _: f'${x:,.0f}'))
fig.savefig(VIZ_DIR / '04_salary_boxplot.png')

# Source: generate_viz.py  -  chart7_roi_bubble()
df = pd.read_csv(OUT_13 / 'roi_summary.csv')
df.columns = ['intervention', 'affected', 'savings', 'budget', 'roi']

fig, ax = plt.subplots(figsize=(12, 8))
for i, (_, row) in enumerate(df.iterrows()):
    ax.scatter(row['savings'], row['roi'],
               s=row['affected'] * 8,
               color=COLORS_10[i], alpha=0.75, edgecolors='white', linewidth=1, zorder=5)
    ax.annotate(row['intervention'],
                (row['savings'], row['roi']),
                textcoords="offset points", xytext=(10, 5), fontsize=10, color=WHITE)

ax.set_xlabel('预计年节省额 (AUD)'); ax.set_ylabel('ROI (节省/投入)')
ax.set_title('NovaCorp 干预措施 ROI 气泡图\n（气泡大小 = 受影响人数）')
ax.xaxis.set_major_formatter(mticker.FuncFormatter(lambda x, _: f'${x/1e6:.1f}M'))
fig.savefig(VIZ_DIR / '07_roi_bubble.png')