FORECASTING PLATFORM

Modules

• Initial forecast: Rolling forecast from start date to end date based on selected time series forecasting model.
• Production forecast: Next period forecast by top model selection based on last forecasting log.
• Feature selection: Select x-series (external features) and their lag periods which are the most correlated for forecasting each y-series.

Time Series Forecasting Model

• Single Exponential Smoothing (Simple Smoothing)
• Double Exponential Smoothing (Holt’s Method)
• Triple Exponential Smoothing (Holt-Winters’ Method)
• Naive (Normal and Seasonal)
• Simple Moving Average
• Weighted Moving Average
• Exponential Moving Average
• ARIMA (Autoregressive Integrated Moving Average)
• ARIMAX (Autoregressive Integrated Moving Average with Explanatory Variable)
• Prophet by Facebook
• Linear Regression
• Random Forest
• XGBoost
• LSTM (Long Short-Term Memory)

Feature Selection Model

• Optimze score by AICc, AIC, Adjusted R-squared

REST API

Features

• Request API to run job module in the background
• Monitor job
• Download output file
• Control job access by user
• Limit maximum job in the platform

API

• Run Module (POST)
• Stop running job (POST)
• List all jobs (POST)
• Get job status (GET)
• Download output (GET)

Deployment

Run without configuration

1. Build Container

git clone [REPO_URL]
cd forecasting-platform
docker build -t [IMAGE_NAME] .

2. Run Container

docker run --name [CONTAINER_NAME] -d -p [PORT]:5000 [IMAGE_NAME]

Run by changing configuration file

1. Build Container

git clone [REPO_URL]
cd forecasting-platform
docker build -t [IMAGE_NAME] .

2. Prepare config.yaml (other directory)

config.yaml

# API key to authenticate while requesting
apikey: [API_KEY]

# Timezone in platform (logfile, database)
timezone: [TIMEZONE]

# Database name (can use from last running container)
database: [JOB_DATABASE_NAME.db]

# Maximum job in the platform to reduce memory of machine
max_job: [N]

# Show debug in webpage or not
debug: [True/False]

# Stream log in container log or not
log_stream: [True/False]

# Number of lines to display in short log in job status
log_lines: [N]

• example of config.yaml file can be found here
• list of all timezone can be found here

3. Run Container

# run without mounting job data
docker run --name [CONTAINER_NAME] \
-v $(pwd)/config.yaml:/app/config.yaml \
-d -p [PORT]:5000 [IMAGE_NAME]

# run by mounting last job database and data
docker run --name [CONTAINER_NAME] \
-v $(pwd)/config.yaml:/app/config.yaml \
-v $(pwd)/tmp:/app/tmp \
-d -p [PORT]:5000 [IMAGE_NAME]