parallel_prophet: A Python repository from iCUE-Solutions

Running multiple timeseries forcasting with Facebook Prophet

This package allows you to run timeseries forecasting in parallel using Facebook Prophet in Google Machine Learning Engine or your local computer.

Usage

The input csv file must look like this:

y	date	id
25	2018-04-24	A
11	2018-04-18	A
7	2017-02-02	A
27	2017-03-19	B
7	2017-04-18	B
10	2017-04-01	B

Where:

y is the target value
date date
id identifier of the timeserie

If you want to use other column names, you must specify them when launching the command.

Running local

python main.py --local_file yes --input_file "../../sample.csv" \
               --index_column "product_uid" \
               --date_column "date" --y_column "q_total" \
               --output_name "../../output.csv" --type "baseline"

The output file contains the results of Facebook Prophet Forecasting

Running on Google Machine Learning Engine

gcloud ml-engine jobs submit training JOB_NAME \
        --module-name=prophet_gcp.main \
        --package-path prophet_gcp/ \
        --region=us-east1 \
        --staging-bucket=gs://your-stagin-bucket/ \
        --scale-tier=BASIC \
        --runtime-version=1.9 \
        -- \
        --input_file gs://path/to/input_file.csv \
        --output_file gs://path/to/output_file/ \
        --index_column id \
        --date_column date \
        --y_column y \
        --type baseline \
        --y_column y \
        --output_name prediction_results.csv \
        --start_date "yyy-MM-dd" \ #only if type is forecast
        --end_date "yyy-MM-dd" \ #only if type is forecast

The output file contains the results of Facebook Prophet Forecasting, they are stored in Google Cloud Storage, the aggregated time series are stored in GCS too with dill.

Using a config file por Prophet

You can optionally pass a configuration file, for both the local and google machine learning version. We provide an example config file called config_example.json.

python main.py --local_file yes --input_file "../../sample.csv" \
               --index_column "product_uid" \
               --date_column "date" --y_column "q_total" \
               --output_name "../../output.csv" --type "baseline \
               --prophet_options config_example.json"

The config file needs to have three keys, though they can be left blank:

prophet_obj_kwds: passed as Prophet(yearly_seasonality=True, **prophet_obj_kwds)
predict_kwds: passed as model.fit(timeseries, **fit_kwds)
fit_kwds: passed as model.predict(time_index, **predict_kwds)

List of arguments for main.py