Prediction API Reference

Base URL

All ML endpoints are prefixed with /v1/ml and typically accessed through the API gateway:

https://api.sgivu.com/v1/ml

For direct access during development:

http://localhost:8000/v1/ml

Authentication

All endpoints require authentication via JWT token or internal service key.

JWT Token (External Clients)

curl -H "Authorization: Bearer YOUR_JWT_TOKEN" \
  https://api.sgivu.com/v1/ml/predict

Internal Service Key

curl -H "X-Internal-Service-Key: YOUR_SERVICE_KEY" \
  http://sgivu-ml:8000/v1/ml/predict

Endpoints

Health Check

/health

GET

Basic health check endpoint without authentication.

Request

GET /health

Response

{
  "status": "ok",
  "version": "v1"
}

Predict Demand

/v1/ml/predict

POST

Generate demand forecast for a specific vehicle segment.

Required Permissions: sgivu.ml.predict or internal service key

Request Schema

{
  "vehicle_type": "CAR",
  "brand": "TOYOTA",
  "model": "COROLLA",
  "line": "XEI 2.0",
  "horizon_months": 6,
  "confidence": 0.95
}

Request Parameters

vehicle_type

string

required

Vehicle category: CAR or MOTORCYCLE

brand

string

required

Vehicle manufacturer (e.g., TOYOTA, HONDA, FORD)

model

string

required

Vehicle model name (e.g., COROLLA, CIVIC, F-150)

line

string

required

Specific vehicle trim/version (e.g., “XEI 2.0”, “EX-L”, “LARIAT”)

The line field is mandatory and cannot be empty. All segments must have a defined line.

horizon_months

integer

default:"6"

Number of months to forecast ahead (1-24)

Minimum: 1
Maximum: 24
Default: 6

confidence

float

default:"0.95"

Confidence level for prediction intervals (0.5-0.99)

Minimum: 0.50 (50%)
Maximum: 0.99 (99%)
Default: 0.95 (95%)
Common values: 0.80, 0.90, 0.95, 0.99

Response Schema

{
  "predictions": [
    {
      "month": "2026-04-01",
      "demand": 45.3,
      "lower_ci": 38.1,
      "upper_ci": 52.5
    },
    {
      "month": "2026-05-01",
      "demand": 47.8,
      "lower_ci": 40.2,
      "upper_ci": 55.4
    }
  ],
  "model_version": "20260306_143022",
  "metrics": {
    "rmse": 3.24,
    "mae": 2.15,
    "mape": 0.087,
    "r2": 0.89,
    "residual_std": 2.8
  }
}

Response Fields

predictions

array

Array of monthly forecast points

Show MonthlyPrediction object

month

string

ISO 8601 date (YYYY-MM-DD) representing the first day of the month

demand

float

Point estimate of expected sales count

lower_ci

float

Lower bound of confidence interval (always ≥ 0)

upper_ci

float

Upper bound of confidence interval

model_version

string

Timestamp-based version identifier of the model used (format: YYYYMMDD_HHMMSS)

metrics

object

Performance metrics from model training

rmse: Root Mean Squared Error
mae: Mean Absolute Error
mape: Mean Absolute Percentage Error
r2: R-squared (coefficient of determination)
residual_std: Standard deviation of residuals (used for CI calculation)

Example Request

curl -X POST https://api.sgivu.com/v1/ml/predict \
  -H "Authorization: Bearer YOUR_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{
    "vehicle_type": "CAR",
    "brand": "CHEVROLET",
    "model": "ONIX",
    "line": "LTZ 1.4",
    "horizon_months": 12,
    "confidence": 0.90
  }'

Predict with History

/v1/ml/predict-with-history

POST

Generate forecast and return historical sales data for visualization.

Required Permissions: sgivu.ml.predict or internal service key

This endpoint is ideal for building charts that show both historical trends and future predictions.

Request Schema

Identical to /predict endpoint (see above).

Response Schema

{
  "predictions": [
    {
      "month": "2026-04-01",
      "demand": 45.3,
      "lower_ci": 38.1,
      "upper_ci": 52.5
    }
  ],
  "history": [
    {
      "month": "2025-01-01",
      "sales_count": 42.0
    },
    {
      "month": "2025-02-01",
      "sales_count": 38.0
    },
    {
      "month": "2025-03-01",
      "sales_count": 44.0
    }
  ],
  "segment": {
    "vehicle_type": "CAR",
    "brand": "CHEVROLET",
    "model": "ONIX",
    "line": "LTZ 1.4"
  },
  "model_version": "20260306_143022",
  "trained_at": "2026-03-06T14:30:22.123456+00:00",
  "metrics": {
    "rmse": 3.24,
    "mae": 2.15,
    "mape": 0.087,
    "r2": 0.89,
    "residual_std": 2.8
  }
}

Additional Response Fields

history

array

Array of historical monthly sales data

Show HistoricalPoint object

month

string

ISO 8601 date (YYYY-MM-DD)

sales_count

float

Actual sales count for that month

segment

object

Normalized segment identifiers used for the prediction

vehicle_type: Canonicalized vehicle type
brand: Canonicalized brand name
model: Canonicalized model name
line: Canonicalized line/trim

trained_at

string

ISO 8601 timestamp when the model was trained

Example Request

curl -X POST https://api.sgivu.com/v1/ml/predict-with-history \
  -H "Authorization: Bearer YOUR_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{
    "vehicle_type": "MOTORCYCLE",
    "brand": "HONDA",
    "model": "CG 160",
    "line": "START",
    "horizon_months": 6,
    "confidence": 0.95
  }'

Retrain Model

/v1/ml/retrain

POST

Trigger model retraining with fresh data from the database.

Required Permissions: sgivu.ml.retrain or internal service key

Retraining is a resource-intensive operation. Use with caution in production environments.

Request Schema

{
  "start_date": "2024-01-01",
  "end_date": "2026-03-06"
}

Request Parameters

start_date

string

Start date for training data (ISO 8601 format: YYYY-MM-DD)If omitted, uses all available historical data.

end_date

string

End date for training data (ISO 8601 format: YYYY-MM-DD)If omitted, uses data up to the current date.

Response Schema

{
  "version": "20260306_150845",
  "metrics": {
    "rmse": 2.89,
    "mae": 1.92,
    "mape": 0.072,
    "r2": 0.91,
    "residual_std": 2.45
  },
  "trained_at": "2026-03-06T15:08:45.678901+00:00",
  "samples": {
    "train": 1847,
    "test": 462,
    "total": 2309
  }
}

Response Fields

version

string

Unique version identifier for the new model (timestamp-based)

metrics

object

Performance metrics from model evaluation

rmse: Lower is better (same units as target variable)
mae: Mean absolute error
mape: Percentage error (0.072 = 7.2%)
r2: Closer to 1.0 is better
residual_std: Used for confidence interval calculations

trained_at

string

ISO 8601 timestamp when training completed

samples

object

Dataset split information

train: Number of samples used for training
test: Number of samples used for evaluation
total: Total samples in dataset

Example Requests

# Retrain with all available data
curl -X POST https://api.sgivu.com/v1/ml/retrain \
  -H "Authorization: Bearer YOUR_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{}'

Get Latest Model

/v1/ml/models/latest

GET

Retrieve metadata for the currently active model.

Required Permissions: sgivu.ml.models or internal service key

Request

GET /v1/ml/models/latest

Response Schema

{
  "version": "20260306_143022",
  "trained_at": "2026-03-06T14:30:22.123456+00:00",
  "target": "sales_count",
  "features": [
    "vehicle_type",
    "brand",
    "model",
    "line",
    "purchases_count",
    "avg_margin",
    "avg_sale_price",
    "avg_purchase_price",
    "avg_days_inventory",
    "inventory_rotation",
    "lag_1",
    "lag_3",
    "lag_6",
    "rolling_mean_3",
    "rolling_mean_6",
    "month",
    "year",
    "month_sin",
    "month_cos"
  ],
  "metrics": {
    "rmse": 3.24,
    "mae": 2.15,
    "mape": 0.087,
    "r2": 0.89,
    "residual_std": 2.8
  },
  "candidates": [
    {
      "model": "linear_regression",
      "rmse": 4.12,
      "mae": 3.05,
      "mape": 0.124,
      "r2": 0.76,
      "samples": 462
    },
    {
      "model": "random_forest",
      "rmse": 3.45,
      "mae": 2.31,
      "mape": 0.095,
      "r2": 0.85,
      "samples": 462
    },
    {
      "model": "xgboost",
      "rmse": 3.24,
      "mae": 2.15,
      "mape": 0.087,
      "r2": 0.89,
      "samples": 462
    }
  ],
  "train_samples": 1847,
  "test_samples": 462,
  "total_samples": 2309
}

Response Fields

version

string

Model version identifier

trained_at

string

Training completion timestamp

target

string

Target variable name (typically sales_count)

features

array

List of feature names used by the model

metrics

object

Performance metrics of the selected best model

candidates

array

Comparison of all evaluated models during trainingShows which algorithm performed best and by how much.

train_samples

integer

Number of samples in training set

test_samples

integer

Number of samples in test set

total_samples

integer

Total dataset size

Example

curl https://api.sgivu.com/v1/ml/models/latest \
  -H "Authorization: Bearer YOUR_TOKEN"

Error Responses

All endpoints follow consistent error response format:

{
  "detail": "Error message describing what went wrong"
}

Common Error Codes

400 Bad Request

Invalid input parameters or missing required fields

{"detail": "Falta la línea del vehículo. Incluye brand/model/line completos para predecir."}

401 Unauthorized

Missing or invalid authentication token

{"detail": "Invalid token or authentication required"}

403 Forbidden

Insufficient permissions for the requested operation

{"detail": "Missing required permission: sgivu.ml.retrain"}

404 Not Found

Requested resource doesn’t exist

{"detail": "No se encontró historial para la combinación solicitada."}

422 Unprocessable Entity

Validation error in request payload

{
  "detail": [
    {
      "loc": ["body", "horizon_months"],
      "msg": "ensure this value is less than or equal to 24",
      "type": "value_error.number.not_le"
    }
  ]
}

500 Internal Server Error

Unexpected server error (check logs)

{"detail": "Aún no existe un modelo entrenado."}

Rate Limiting

Rate limiting is typically handled at the API gateway level. Check gateway documentation for specific limits.

OpenAPI Specification

The complete OpenAPI 3.0 specification is available at:

GET /openapi.json

You can import this into Postman, Insomnia, or other API clients for automatic request generation.

SDK Examples

For more comprehensive examples and client library usage, see:

Training Process - Feature engineering and model training details
Model Management - Versioning and lifecycle management
Frontend Integration - Building client applications

Next Steps

Try the API

Test endpoints using the interactive docs at /docs

Understanding Predictions

Learn how forecasts are generated

Model Versioning

Manage and compare model versions

Authentication

Configure JWT and permissions

ML Service

​Base URL

​Authentication

​JWT Token (External Clients)

​Internal Service Key

​Endpoints

​Health Check

​Request

​Response

​Predict Demand

​Request Schema

​Request Parameters

​Response Schema

​Response Fields

​Example Request

​Predict with History

​Request Schema

​Response Schema

​Additional Response Fields

​Example Request

​Retrain Model

​Request Schema

​Request Parameters

​Response Schema

​Response Fields

​Example Requests

​Get Latest Model

​Request

​Response Schema

​Response Fields

​Example

​Error Responses

​Common Error Codes

​Rate Limiting

​OpenAPI Specification

​SDK Examples

​Next Steps

Try the API

Understanding Predictions

Model Versioning

Authentication

Base URL

Authentication

JWT Token (External Clients)

Internal Service Key

Endpoints

Health Check

Request

Response

Predict Demand

Request Schema

Request Parameters

Response Schema

Response Fields

Example Request

Predict with History

Request Schema

Response Schema

Additional Response Fields

Example Request

Retrain Model

Request Schema

Request Parameters

Response Schema

Response Fields

Example Requests

Get Latest Model

Request

Response Schema

Response Fields

Example

Error Responses

Common Error Codes

Rate Limiting

OpenAPI Specification

SDK Examples

Next Steps