text stringlengths 0 81 |
|---|
# train_model.py |
# Generates a synthetic dataset, trains an ML model, and saves cost_estimator.pkl |
import numpy as np |
import pandas as pd |
from sklearn.ensemble import RandomForestRegressor |
from sklearn.preprocessing import OneHotEncoder |
from sklearn.compose import ColumnTransformer |
from sklearn.pipeline import Pipeline |
import joblib |
np.random.seed(42) |
# Settings |
n_samples = 800 |
project_types = ["Residential", "Commercial", "Infrastructure", "Road"] |
location_opts = ["Urban", "Semi-Urban", "Rural"] |
material_opts = ["Low", "Medium", "High"] |
# base rates (USD per m2) for synthetic generation |
base_rates = { |
"Residential": 350, |
"Commercial": 500, |
"Infrastructure": 700, |
"Road": 400 |
} |
material_factor = {"Low": 0.9, "Medium": 1.0, "High": 1.2} |
location_factor = {"Urban": 1.1, "Semi-Urban": 1.0, "Rural": 0.85} |
rows = [] |
for _ in range(n_samples): |
pt = np.random.choice(project_types) |
loc = np.random.choice(location_opts) |
mat = np.random.choice(material_opts) |
# area ranges by project type |
if pt == "Residential": |
area = int(np.random.uniform(50, 400)) |
elif pt == "Commercial": |
area = int(np.random.uniform(200, 3000)) |
elif pt == "Infrastructure": |
area = int(np.random.uniform(500, 5000)) |
else: # Road |
area = int(np.random.uniform(1000, 10000)) |
# generate cost with some noise |
base = base_rates[pt] |
cost = area * base * material_factor[mat] * location_factor[loc] |
noise = np.random.normal(loc=0.0, scale=0.08 * cost) # 8% noise |
cost = max(100.0, cost + noise) |
rows.append({ |
"Project_Type": pt, |
"Area_m2": area, |
"Location": loc, |
"Material_Quality": mat, |
"Cost_USD": round(cost, 2) |
}) |
df = pd.DataFrame(rows) |
print("Sample data:") |
print(df.head()) |
# train/test split (simple) |
X = df.drop(columns=["Cost_USD"]) |
y = df["Cost_USD"] |
categorical = ["Project_Type", "Location", "Material_Quality"] |
numeric = ["Area_m2"] |
preprocessor = ColumnTransformer([ |
("cat", OneHotEncoder(handle_unknown="ignore"), categorical), |
("num", "passthrough", numeric) |
]) |
pipeline = Pipeline([ |
("preproc", preprocessor), |
("rf", RandomForestRegressor(n_estimators=100, random_state=42)) |
]) |
print("Training model...") |
pipeline.fit(X, y) |
# save |
joblib.dump(pipeline, "cost_estimator.pkl") |
print("Saved model to cost_estimator.pkl") |
No dataset card yet
- Downloads last month
- 3