Spaces:
Sleeping
Sleeping
File size: 6,974 Bytes
1372d99 42dfbed 1372d99 42dfbed | 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 | """
Ce module contient les fonctions de calcul du PSI (Population Stability Index) pour les variables d'entrée utilisées dans le modèle de prédiction de rendement agricole.
Il inclut :
- get_input_drift_summary() : Calcule un résumé des statistiques descriptives pour les variables d'entrée à partir des données de production.
- get_numeric_psi() : Calcule le PSI pour les variables d'entrée numériques en comparant la distribution de référence d'entraînement avec la distribution des données reçues en production.
- get_distribution_comparison(feature_name) : Compare la distribution d'une variable d'entrée spécifique entre
la référence d'entraînement et les données reçues en production, et retourne les valeurs pour les deux distributions.
"""
import math
from src.api.monitoring.db import get_connection
def get_input_drift_summary():
"""Calcule un résumé des statistiques descriptives pour les variables d'entrée
à partir des données de production.
Returns:
dict: Un dictionnaire contenant les statistiques descriptives pour chaque variable d'entrée.
"""
query = """
SELECT
input_json
FROM api_requests
WHERE input_json IS NOT NULL;
"""
with get_connection() as conn:
with conn.cursor() as cursor:
cursor.execute(query)
rows = cursor.fetchall()
inputs = [row[0] for row in rows]
numeric_fields = ["temp", "rainfall_mm", "pesticides", "trend"]
summary = {}
for field in numeric_fields:
values = [
float(input_data[field])
for input_data in inputs
if field in input_data and input_data[field] is not None
]
if values:
summary[field] = {
"count": len(values),
"mean": round(sum(values) / len(values), 2),
"min": round(min(values), 2),
"max": round(max(values), 2),
}
else:
summary[field] = {
"count": 0,
"mean": None,
"min": None,
"max": None,
}
return summary
import math
def _calculate_psi(reference_values, production_values, bins=10):
"""
Calcule le PSI entre une distribution de référence
et une distribution de production.
"""
if not reference_values or not production_values:
return None
min_value = min(reference_values)
max_value = max(reference_values)
if min_value == max_value:
return 0
bin_width = (max_value - min_value) / bins
psi = 0
for i in range(bins):
lower = min_value + i * bin_width
upper = min_value + (i + 1) * bin_width
if i == bins - 1:
ref_count = sum(lower <= value <= upper for value in reference_values)
prod_count = sum(lower <= value <= upper for value in production_values)
else:
ref_count = sum(lower <= value < upper for value in reference_values)
prod_count = sum(lower <= value < upper for value in production_values)
ref_pct = ref_count / len(reference_values)
prod_pct = prod_count / len(production_values)
ref_pct = max(ref_pct, 0.0001)
prod_pct = max(prod_pct, 0.0001)
psi += (prod_pct - ref_pct) * math.log(prod_pct / ref_pct)
return round(psi, 4)
def _get_psi_status(psi):
if psi is None:
return "Données insuffisantes"
if psi < 0.1:
return "Stable"
if psi < 0.2:
return "Légère dérive"
return "Drift important"
def get_numeric_psi():
features = ["temp", "rainfall_mm", "pesticides", "trend"]
results = []
with get_connection() as conn:
with conn.cursor() as cursor:
for feature in features:
cursor.execute(
"""
SELECT feature_value
FROM reference_distributions
WHERE feature_name = %s
AND source = 'train';
""",
(feature,)
)
reference_values = [
float(row[0])
for row in cursor.fetchall()
if row[0] is not None
]
cursor.execute(
"""
SELECT input_json
FROM api_requests
WHERE input_json IS NOT NULL;
"""
)
production_inputs = cursor.fetchall()
production_values = []
for row in production_inputs:
input_data = row[0]
if feature in input_data and input_data[feature] is not None:
production_values.append(float(input_data[feature]))
psi = _calculate_psi(reference_values, production_values)
results.append(
{
"feature": feature,
"psi": psi,
"status": _get_psi_status(psi),
"reference_count": len(reference_values),
"production_count": len(production_values),
}
)
return results
def get_distribution_comparison(feature_name):
"""
Compare la distribution d'une variable entre :
- la référence d'entraînement
- les données reçues en production
"""
allowed_features = [
"temp",
"rainfall_mm",
"pesticides",
"trend",
]
if feature_name not in allowed_features:
raise ValueError(
f"Variable non autorisée : {feature_name}"
)
with get_connection() as conn:
with conn.cursor() as cursor:
cursor.execute(
"""
SELECT feature_value
FROM reference_distributions
WHERE feature_name = %s
AND source = 'train';
""",
(feature_name,)
)
reference_values = [
float(row[0])
for row in cursor.fetchall()
if row[0] is not None
]
cursor.execute(
"""
SELECT input_json
FROM api_requests
WHERE input_json IS NOT NULL;
"""
)
production_inputs = cursor.fetchall()
production_values = []
for row in production_inputs:
input_data = row[0]
if (
feature_name in input_data
and input_data[feature_name] is not None
):
production_values.append(
float(input_data[feature_name])
)
return {
"feature": feature_name,
"reference": reference_values,
"production": production_values,
} |