Material-Prediction-API-Service

Product Price Forecasting API Service

This API service provides a price forecasting solution for a product based on the contributing materials that form the final product. The service uses machine learning models (specifically NeuralProphet) to predict the future price of materials over a defined period and returns a forecast of the product price based on the weights and spot prices of various input materials.

Table of Contents

• Description
• Features
• Endpoints
  • Status
  • Help
  • Calculate
• Running the Service
• How the API Works

---

Description

This service accepts multiple material types as inputs, such as weight (in KG), spot price, and forecast period (in months), and returns the predicted price of the product over the specified period. The forecast starts from February 2023, as that is the last timestamp in the training data.

Key Materials Used in the Forecast:

• ST37 (Spot price, weight)
• Copper (Spot price, weight)
• High Carbon (Spot price, weight)
• Medium Carbon (Spot price, weight)
• Aluminum (Weight)
• Grey Cast Iron (Spot price, weight)
• Nodular Cast Iron (Spot price, weight)
• Nonalloy Cast (Spot price, weight)
• Labor Hours

Users can also specify a forecasting period (months) to define how many months into the future the predictions should go (defaults to 24 months).

Features

• Flexible Input: Accepts a wide range of materials and prices.
• Custom Forecast Period: Allows the user to specify how many months to predict into the future.
• Validation: Ensures all required inputs are provided and that no price is missing for any supplied material.
• NeuralProphet Models: Uses time-series forecasting models for accurate predictions.

Endpoints

1. GET /status

Checks if the API service is running.

Example Request:

http://innosale.sagresearch.de:8012/status

Example Response:

{
  "Status": "Service Running"
}

2. GET /help

Provides information on the materials and arguments accepted by the API service.

Example Request:

http://innosale.sagresearch.de:8012/help

Example Response:

{
  "st37": "Weight of ST37 in KG",
  "p_st37": "Spot price of ST37",
  ...
  "months": "Forecasting period in months (default is 24 months)"
}

3. GET /calculate

Returns the predicted product price over the forecast period based on the input materials and prices.

Parameters:

• st37: Weight of ST37 in KG
• copper: Weight of Copper in KG
• p_st37: Spot price of ST37
• p_high_carbon: Spot price of High Carbon
• alu: Weight of Aluminum in KG
• labour: Labor hours
• high_carbon: Weight of High Carbon in KG
• medium_carbon: Weight of Medium Carbon in KG
• p_medium_carbon: Spot Price of Medium Carbon
• p_nodular_cast_iron: Spot Price of Nodular Cast Iron
• nodular_cast_iron: Weight of Nodular Cast Iron in KG
• grey_cast_iron: Weight of Grey Cast Iron in KG
• p_grey_cast_iron: Spot Price of Grey Cast Iron
• nonalloy_cast: Weight of Nonalloy Cast in KG
• p_nonalloy_cast: Spot Price of Nonalloy Cast
• months: Forecasting period in months (optional, default is 24 months)

Example Request:

http://innosale.sagresearch.de:8012/calculate?st37=100&p_st37=1500&alu=50&labour=10&months=12

Example Response:

  {
    "ds": "2023-03",
    "total_product_value": 52300.12
  },
  {
    "ds": "2023-04",
    "total_product_value": 52045.20
  },
  ...

Running the Service

Requirements:

• Python 3.8+
• FastAPI
• NeuralProphet
• Pandas

Installation:

1. Clone the repository:

git clone https://github.com/Mahmoud-1995/Material-Prediction-API-Service.git

2. Install the required dependencies:

pip install -r requirements.txt

3. Start the FastAPI server::

python3.8 -m uvicorn main:app --host 10.0.11.20(host ip) --port 8080(port number)

How the API Works

1- Input Validation: The API checks if the inputs provided have corresponding values for both weight and spot price. If any spot price is missing while its weight is provided, or vice versa, the API will return a validation error.

2- Prediction: The API leverages pre-trained NeuralProphet models to forecast the price of each material over the specified forecast period. The total price is calculated by multiplying the material weight with the forecasted price for that period.

3- Output: The API returns a list of predicted product prices for each month in the forecast period, formatted as a time-series.