Spaces:
Sleeping
Sleeping
File size: 19,018 Bytes
ceb3888 89b0184 ceb3888 89b0184 ceb3888 89b0184 ceb3888 89b0184 bb6bf65 ceb3888 89b0184 ceb3888 89b0184 ceb3888 89b0184 ceb3888 89b0184 ceb3888 89b0184 ceb3888 89b0184 ceb3888 89b0184 ceb3888 89b0184 ceb3888 89b0184 ceb3888 89b0184 ceb3888 89b0184 ceb3888 89b0184 ceb3888 89b0184 ceb3888 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 | import streamlit as st
import pandas as pd
import json
import plotly.express as px
from pathlib import Path
import plotly.graph_objects as go
st.set_page_config(page_title="Monitoring Drift", layout="wide")
# Charger le rapport Evidently
with open("Monitoring/drift_report.json", "r", encoding="utf-8") as f:
drift_data = json.load(f)
# # Tests pour vérifier la structure du json drift
# for m in drift_data["metrics"]:
# if m["metric"] == "ColumnSummaryMetric":
# st.json(m)
# break
# for m in drift_data["metrics"]:
# if m["metric"] == "ColumnSummaryMetric":
# st.write(m)
# break
# for m in drift_data["metrics"]:
# if m["metric"] == "ColumnSummaryMetric":
# print("======")
# print(m.keys())
# print(m["result"].keys())
# break
# for m in drift_data["metrics"]:
# if m["metric"] == "ColumnDriftMetric":
# st.json(m)
# break
# metrics_names = set(
# m["metric"]
# for m in drift_data["metrics"])
# st.write(metrics_names)
# st.json(drift_data["metrics"][0])
# Charger les logs bruts
df_logs = pd.read_parquet("logs/predictions_log.parquet")
df_logs["timestamp"] = pd.to_datetime(df_logs["timestamp_x"], unit="s")
# Charger les logs après optimisation
df_logs_opti = pd.read_parquet("logs/predictions_log_batch.parquet")
df_logs_opti["timestamp_model"] = pd.to_datetime(df_logs_opti["timestamp_model"], unit="s")
# st.write(df_logs.head())
# st.write(df_logs_opti.head())
st.title("📊 Monitoring du Drift — Dashboard MLOps")
# ONGLET 1 — Vue d’ensemble
tab1, tab2, tab3, tab4, tab5 = st.tabs(["VUE D'ENSEMBLE", "DERIVE PAR COLONNE", "LOGS BRUTS", "IMPACT OPTIMISATION", "COMPARAISON UNITAIRE vs BATCH"])
with tab1:
# Colonnes en dérive - Résultat global
dataset_drift = next(
m for m in drift_data["metrics"]
if m["metric"] == "DatasetDriftMetric")
drift_res = dataset_drift["result"]
with st.container():
st.markdown("""<p style='font-size: 22px; font-style: italic;'>Performance Globale Colonnes</p>""", unsafe_allow_html=True)
col1, col2, col3, col4 = st.columns([1,1,1,1])
st.markdown("<br>", unsafe_allow_html=True)
col1.metric("Prédictions", len(df_logs))
col2.metric("Colonnes analysées", drift_res["number_of_columns"])
col3.metric("Colonnes en dérive", drift_res["number_of_drifted_columns"])
col4.metric( "Taux de dérive", f"{100*drift_res['share_of_drifted_columns']:.1f}%")
# KPI API
with st.container():
st.markdown("""<p style='font-size: 22px; font-style: italic;'>Performance API</p>""", unsafe_allow_html=True)
c1, c2, c3, c4, c5, c6 = st.columns([1,1,1,1,1,1])
st.markdown("<br>", unsafe_allow_html=True)
c1.metric("Latence moyenne", f"{df_logs['latency_ms'].mean():.2f}")
c2.metric("Temps d'inférence moyen", f"{df_logs['inference_ms_x'].mean():.2f}")
c3.metric("CPU moyen", f"{df_logs['cpu_percent'].mean():.1f}%")
c4.metric("RAM moyenne", f"{df_logs['ram_percent'].mean():.1f}%")
c5.metric("Charge système", f"{df_logs['system_load'].mean():.2f}")
c6.metric("Threads moyens", f"{df_logs['num_threads'].mean():.0f}")
# KPI Métier
with st.container():
st.markdown("""<p style='font-size: 22px; font-style: italic;'>Performance Métier</p>""", unsafe_allow_html=True)
m1, m2, m3 = st.columns([1,1,1])
st.markdown("<br>", unsafe_allow_html=True)
m1.metric("Probalité moyenne", f"{df_logs['score_x'].mean():.2f}")
m2.metric("Seuil optimal moyen", f"{df_logs['threshold_x'].mean():.3f}")
m3.metric("Décision majoritaire", df_logs["decision_x"].mode()[0])
with st.container():
g1, g2 = st.columns([1,1])
# Graphique distribution des scores
score_df = (df_logs["score_x"].value_counts().reset_index())
fig = px.bar(score_df, x = "score_x", y = "count", title = "Répartition des scores")
g1.plotly_chart(fig, width = "stretch")
# Graphique distribution décisions
decision_df = (df_logs["decision_x"].value_counts().reset_index())
fig = px.bar(decision_df, x = "decision_x", y = "count", title = "Répartition des décisions")
g2.plotly_chart(fig, width = "stretch")
with st.container():
h1, h2, h3 = st.columns([1,1,1])
# Latence dans le temps
fig = px.line(df_logs, x="timestamp", y="latency_ms", title = "Distribution de la latence dans le temps")
h1.plotly_chart(fig, width = "stretch")
# CPU dans le temps
fig = px.line(df_logs, x="timestamp", y="cpu_percent", title = "Distribution du CPU dans le temps")
h2.plotly_chart(fig, width = "stretch")
# RAM dans le temps
fig = px.line(df_logs, x="timestamp", y="ram_percent", title = "Distribution de la RAM dans le temps")
h3.plotly_chart(fig, width = "stretch")
# Graphique distribution colonnes en dérive
drift_graph = pd.DataFrame({
"Etat": [
"En dérive",
"Sans dérive"],
"Nombre": [
drift_res["number_of_drifted_columns"],
drift_res["number_of_columns"]
- drift_res["number_of_drifted_columns"]]})
fig = px.bar(drift_graph, x = "Etat", y = "Nombre", title = "Répartition des colonnes en dérive")
st.plotly_chart(fig, width = "stretch")
# Liste des colonnes en drift
drifted_cols = []
for m in drift_data["metrics"]:
if m["metric"] != "ColumnDriftMetric":
continue
if m['result']['drift_detected']:
drifted_cols.append(m["result"]["column_name"])
st.subheader("Colonnes en dérive")
st.write(drifted_cols)
# ONGLET 2 — Dérive par colonne
with tab2:
st.header("📈 Dérive par colonne")
summary_dict = {}
for m in drift_data["metrics"]:
if m["metric"] != "ColumnSummaryMetric":
continue
result = m['result']
summary_dict[result["column_name"]] = {
"missing": result["current_characteristics"]["missing"],
"missing_pct": result["current_characteristics"]["missing_percentage"],
"unique": result["current_characteristics"]["unique"]}
drift_rows = []
for m in drift_data["metrics"]:
if m["metric"] != "ColumnDriftMetric":
continue
result = m['result']
col = result["column_name"]
# missing_data = (summary_dict[col]["current_characteristics"]["missing"])
drift_rows.append({
"Colonne": col,
"Type": result["column_type"],
"Distance drift": result["drift_score"],
"Dérive(%)": round(result["drift_score"] * 100, 2),
"Seuil(%)": round(result["stattest_threshold"] * 100, 2),
"Drift détecté": result["drift_detected"],
"Test": result["stattest_name"],
"Nombre valeurs manquantes": summary_dict[col]["missing"],
"Taux valeurs manquantes": summary_dict[col]["missing_pct"],
"Nombre de valeurs uniques": summary_dict[col]["unique"]})
df_drift = pd.DataFrame(drift_rows)
df_drift_filtered = (df_drift[df_drift["Drift détecté"] == True].drop_duplicates(subset=["Colonne"])) # uniquement les colonnes en dérive pour voir le taux de dérive par colonne
st.dataframe(df_drift_filtered.sort_values("Distance drift", ascending = False), width = "stretch")
selected_col = st.selectbox("Choisir une colonne", df_drift_filtered["Colonne"], key = "drift_column")
col_stats = df_drift_filtered[df_drift_filtered["Colonne"] == selected_col].iloc[0]
metric = next(m for m in drift_data["metrics"]if (m["metric"] == "ColumnDriftMetric" and m["result"]["column_name"] == selected_col))
x_ref = metric["result"]["reference"]["small_distribution"]["x"]
y_ref = metric["result"]["reference"]["small_distribution"]["y"]
x_cur = metric["result"]["current"]["small_distribution"]["x"]
y_cur = metric["result"]["current"]["small_distribution"]["y"]
summary_metric = next(m for m in drift_data["metrics"] if (m["metric"] == "ColumnSummaryMetric" and m["result"]["column_name"] == selected_col))
missing_ref = summary_metric["result"]["reference_characteristics"]["missing"]
missing_cur = summary_metric["result"]["current_characteristics"]["missing"]
with st.container():
st.write(f"Analyse détaillée: {selected_col}")
st.markdown(f"""
- **Distance drift** : `{col_stats['Distance drift']:.4f}`
- **Dérive (%)** : `{col_stats['Dérive(%)']} %`
- **Seuil (%)** : `{col_stats['Seuil(%)']} %`
- **Drift détecté** : `{"Oui" if col_stats['Drift détecté'] else "Non"}`
""")
d1, d2 = st.columns([1,1])
# Distribution des données
fig = go.Figure()
fig.add_bar(x = x_ref, y = y_ref, name = "Référence")
fig.add_bar(x = x_cur, y = y_cur, name = "Production")
fig.update_layout(barmode = "group", title = f"Distribution - {selected_col}", xaxis_title = "Valeurs", yaxis_title = "Densité", legend_title = "")
d1.plotly_chart(fig, width = "stretch")
# Distribution valeurs manquantes
fig_missing = go.Figure()
fig_missing.add_bar(x = ["Référence"], y = [missing_ref], name = "Référence")
fig_missing.add_bar(x = ["Production"], y = [missing_cur], name = "Production")
fig_missing.update_layout(barmode="group", title=f"Valeurs manquantes - {selected_col}", yaxis_title = "Nombre de valeurs manquantes")
d2.plotly_chart(fig_missing, width = "stretch")
# ONGLET 3 — Résumé par colonne (finalement faire plutôt un tableau?)
# with tab3:
# st.header("📘 Résumé statistique par colonne")
# summaries = {}
# for m in drift_data["metrics"]:
# if m["metric"] != "ColumnSummaryMetric":
# continue
# result2 = m['result']
# if ("column_name" not in result2 or result2["column_name"] is None or "reference_characteristics" not in result2 or "current_characteristics" not in result2):
# continue
# summaries[result2["column_name"]] = result2
# options = sorted(summaries.keys())
# col_selectione = st.selectbox("Choisir une colonne", options, index=0)
# # col_selectione = st.selectbox("Choisir une colonne", sorted(summaries.keys()), key="summary_column")
# summary = summaries[col_selectione]
# ref = summary["reference_characteristics"]
# cur = summary["current_characteristics"]
# st.subheader(f"🔹 {col_selectione}")
# col1, col2 = st.columns(2)
# with col1:
# st.markdown("### Référence")
# st.dataframe(pd.DataFrame(ref.items(), columns=["Métrique", "Valeur"]), width = True)
# with col2:
# st.markdown("### Production")
# st.dataframe(pd.DataFrame(cur.items(), columns=["Métrique", "Valeur"]), width = True)
# # comparaison synthétique
# compare_df = pd.DataFrame({
# "Métrique": ["count", "missing", "mean", "std", "min", "p25", "p50", "p75", "max", "unique"],
# "Référence": [ref.get("count"), ref.get("missing"), ref.get("mean"), ref.get("std"), ref.get("min"), ref.get("p25"),
# ref.get("p50"), ref.get("p75"), ref.get("max"), ref.get("unique")],
# "Production": [cur.get("count"), cur.get("missing"), cur.get("mean"), cur.get("std"), cur.get("min"), cur.get("p25"),
# cur.get("p50"), cur.get("p75"), cur.get("max"), cur.get("unique")]})
# st.markdown("### Comparaison directe")
# st.dataframe(compare_df, width = True)
# ONGLET 4 — Logs bruts
with tab3:
st.header("📄 Logs bruts (production)")
st.dataframe(df_logs.head(200))
col = st.selectbox("Visualiser une colonne :", df_logs.columns)
fig = px.histogram(df_logs, x=col, nbins=30, title=f"Distribution de {col}")
st.plotly_chart(fig, width="stretch")
# ONGLET 5 - KPI API après optimisation du processing
with tab4:
# KPI API
with st.container():
st.markdown("""<p style='font-size: 22px; font-style: italic;'>Performance API (Batch)</p>""", unsafe_allow_html=True)
if df_logs_opti.empty:
st.warning("Aucun batch n'a encore été exécuté. Lance un batch pour voir les KPI.")
else:
k1, k2, k3, k4, k5, k6 = st.columns([1,1,1,1,1,1])
k1.metric("Latence moyenne", f"{df_logs_opti['latency_ms_ops'].mean():.2f}")
k2.metric("Temps d'inférence moyen", f"{df_logs_opti['inference_ms_model'].mean():.2f}")
k3.metric("CPU moyen", f"{df_logs_opti['cpu_percent_ops'].mean():.1f}%")
k4.metric("RAM moyenne", f"{df_logs_opti['ram_percent_ops'].mean():.1f}%")
k5.metric("Charge système", f"{df_logs_opti['system_load_ops'].mean():.2f}")
k6.metric("Threads moyens", f"{df_logs_opti['num_threads_ops'].mean():.0f}")
st.markdown("<br>", unsafe_allow_html=True)
with st.container():
o1, o2, o3 = st.columns([1,1,1])
# Latence dans le temps
fig = px.line(df_logs_opti, x="timestamp_model", y="latency_ms_ops", title = "Latence du batch dans le temps")
o1.plotly_chart(fig, width = "stretch")
# CPU dans le temps
fig = px.line(df_logs_opti, x="timestamp_model", y="cpu_percent_ops", title = "CPU utilisé par le batch")
o2.plotly_chart(fig, width = "stretch")
# RAM dans le temps
fig = px.line(df_logs_opti, x="timestamp_model", y="ram_percent_ops", title = "RAM dutilisé par le batch")
o3.plotly_chart(fig, width = "stretch")
fig = px.scatter(df_logs_opti, x = "batch_size_model", y = "latency_ms_ops", trendline = "ols", title = "Relation entre batch_size et latence")
st.plotly_chart(fig, use_container_width=True)
# ONGLET 6 - Compaaison des kpis avant et après optimisation
with tab5:
if df_logs.empty or df_logs_opti.empty:
st.warning("Exécute au moins une requête unitaire et un batch pour afficher la comparaison.")
else:
# Calculs
df_unit = df_logs.copy()
df_batch = df_logs_opti.copy()
# Latence par client
df_unit["latence_par_client"] = df_unit["latency_ms"]
df_batch["latence_par_client"] = df_batch["latency_ms_ops"] / df_batch["batch_size_model"]
# Inférence par client
df_unit["inference_par_client"] = df_unit["inference_ms_x"]
df_batch["inference_par_client"] = df_batch["inference_ms_model"] / df_batch["batch_size_model"]
# CPU / RAM / Load / Threads
df_unit["cpu"] = df_unit["cpu_percent"]
df_batch["cpu"] = df_batch["cpu_percent_ops"]
df_unit["ram"] = df_unit["ram_percent"]
df_batch["ram"] = df_batch["ram_percent_ops"]
df_unit["load"] = df_unit["system_load"]
df_batch["load"] = df_batch["system_load_ops"]
df_unit["threads"] = df_unit["num_threads"]
df_batch["threads"] = df_batch["num_threads_ops"]
# KPI
st.subheader("Performances par client")
p1, p2, p3 = st.columns([1,1,1])
# st.markdown("""<p style='font-size: 22px; font-style: italic;'>Latence par client</p>""", unsafe_allow_html=True)
p1.metric("Latence par client (unitaire)", f"{df_unit['latence_par_client'].mean():.2f} ms")
p2.metric("Latence par client (batch)", f"{df_batch['latence_par_client'].mean():.2f} ms")
p3.metric("Gain", f"{df_unit['latence_par_client'].mean() / df_batch['latence_par_client'].mean():.1f}")
# st.markdown("<br>", unsafe_allow_html=True)
i1, i2, i3 = st.columns([1,1,1])
# st.markdown("""<p style='font-size: 22px; font-style: italic;'>Inférence par client</p>""", unsafe_allow_html=True)
i1.metric("Inférence par client (unitaire)", f"{df_unit['inference_par_client'].mean():.2f} ms")
i2.metric("Inférence par client (batch)", f"{df_batch['inference_par_client'].mean():.2f} ms")
i3.metric("Gain", f"{df_unit['inference_par_client'].mean() / df_batch['inference_par_client'].mean():.1f}")
st.markdown("<br>", unsafe_allow_html=True)
# kpi CPU/RAM/LOAD/THREADS
st.subheader("Charge système")
s1, s2, s3, s4 = st.columns([1,1,1,1])
s1.metric("CPU (%) unitaire", f"{df_unit['cpu'].mean():.1f}%")
s2.metric("CPU (%) batch", f"{df_batch['cpu'].mean():.1f}%")
s3.metric("RAM (%) unitaire", f"{df_unit['ram'].mean():.1f}%")
s4.metric("RAM (%) batch", f"{df_batch['ram'].mean():.1f}%")
s5, s6, s7, s8 = st.columns(4)
s5.metric("Load unitaire", f"{df_unit['load'].mean():.2f}")
s6.metric("Load batch", f"{df_batch['load'].mean():.2f}")
s7.metric("Threads unitaire", f"{df_unit['threads'].mean():.0f}")
s8.metric("Threads batch", f"{df_batch['threads'].mean():.0f}")
st.markdown("<hr>", unsafe_allow_html=True)
# Graphiques comparatifs
st.subheader("Visualisations comparatives")
c1, c2 = st.columns(2)
fig = px.box(pd.concat([df_unit.assign(mode = "Unitaire"), df_batch.assign(mode = "Batch")]), x = "mode", y = "latence_par_client", title = "Latence par client : Unitaire vs Batch")
c1.plotly_chart(fig, use_container_width=True)
fig = px.box(pd.concat([df_unit.assign(mode = "Unitaire"), df_batch.assign(mode = "Batch")]), x = "mode", y = "inference_par_client", title = "Inférence par client : Unitaire vs Batch")
c2.plotly_chart(fig, use_container_width=True)
st.markdown("<br>", unsafe_allow_html=True)
# CPU
fig = px.box(pd.concat([df_unit.assign(mode = "Unitaire"), df_batch.assign(mode = "Batch")]), x = "mode", y = "cpu", title = "CPU (%) : Unitaire vs Batch")
st.plotly_chart(fig, use_container_width=True)
st.markdown("<br>", unsafe_allow_html=True)
# RAM
fig = px.box(pd.concat([df_unit.assign(mode = "Unitaire"), df_batch.assign(mode="Batch")]), x = "mode", y = "ram", title = "RAM (%) : Unitaire vs Batch")
st.plotly_chart(fig, use_container_width=True)
st.markdown("<br>", unsafe_allow_html=True)
# Load
fig = px.box(pd.concat([df_unit.assign(mode = "Unitaire"), df_batch.assign(mode = "Batch")]), x = "mode", y = "load", title = "Charge système : Unitaire vs Batch")
st.plotly_chart(fig, use_container_width=True)
st.markdown("<br>", unsafe_allow_html=True)
# Threads
fig = px.box(pd.concat([df_unit.assign(mode = "Unitaire"), df_batch.assign(mode = "Batch")]), x = "mode", y = "threads", title = "Threads : Unitaire vs Batch")
st.plotly_chart(fig, use_container_width=True)
st.markdown("<br>", unsafe_allow_html=True) |