Time series analysis comprises methods for analyzing time series data in order to extract meaningful statistics and other characteristics of the data. Time series allows you to analyze major patterns such as trends, seasonality, cyclicity, and irregularity. Time series forecasting is the use of a model to predict future values based on previously observed values. Time series are widely used for non-stationary data, like economic, weather, stock price, and retail sales.
This tutorial demonstrates how to quickly build a time-series model using FP Predict Plus operator with a dataset that shows the how the prices differ over a period of time at a paricular retail store outlet.
- Users will learn how to operate with Red Hat Marketplace FP Predict Plus Operator
- Setup a Time-Series Model on FP Predict Plus
- Learn how to configure the Training and Forecast parameters
- View and analyze the time-series results
Please refer to this link for getting started with installation and setup of FP Predict plus operator.
Completing this tutorial should take about 40 minutes.
- Login to your FP Predict Plus Operator by launching it.
- Click on
+ Startbutton
- A prompt will pop-up asking if your data contains Time-Stamp. Select
Yes.
- Another prompt will pop-up asking if you will be predicting values. Select
Yes.
-
Download the dataset train.csv
-
Download the dataset test.csv
-
Enter the details as follows-
- Enter
Job NameasTS(Pls avoid special characters in the name) - Dataset location must be
local - Daily Interval set as
Daily - Tasks
Model + Forecast - Upload training file from the downloaded file
train.csv - Set Target Variable as
item_price
- Upload the forecast file from the downloaded file
test.csv - Set Timestamp variable as
date - Set Timestamp Format as
dd/mm/yyyy - Click on
Run
- Enter
*** Note: It will take about 10 minutes to setup the model, pls wait and don't refresh the page until it's over ****
After the completion you will receive an output as given below-
The above graph shows the time-series of prices over the time period 2013 to 2015.
Click on the analytics icon to analyze your results
- The first table on the top left corner, depcits the actual v/s predicted and the difference in prediction for each time interval. Note that it provides only the head values of the result table
- The graph on the top right represents the same output as received on the dashboard
- The
Modeling MetricsandForecast Metricsprovides measures of the trained model using techniques such as Mean Error, Root Mean Square,Mean Percentage Error, Mean Absolute Percentage Error and Mean Absolute Scaled Error. These will allow us to evaluate how well the underlying model is trained and forecasts. - Log table provides metadata of the trained file, such as the number of rows and time taken to train.
- The last table is about the
Important Rowssince ours is a time-series model with only one underlying feature
distribution.
-
Must have only 2 other columns apart from the date variable which are Row sequence number and Target Variable.
-
Ensure the time interval is correctly maintained through the training and forecast data. Eg: If you put the time interval as Daily, your dataset MUST contain only one record for a particular day.
-
Additionally, if you set Daily as interval, your forecast data MUST have a forecast for everyday without a gap.
-
If you encounter the below error-
Clear your browser cache or try in another browser.
For further reference, look at the datasets used in this tutorial.
In summary, this tutorial helps you to understand how to perform time-series analysis using the FP Predict Plus Operator hosted on Red Hat Market Place.