Getting Started
Examining the analysis workflow.
Welcome to Correlation Studio
This is an introductory guide to using the application's Creator workflow. You'll find the most common usage path and some tips & tricks here. We're building in public so this is a living document.
1. Dataset Wizard Your journey starts and ends with data. That's what we're here for. Correlation Studio provides four modalities for ingesting data:
1. Local Files (CSV, Excel) 2. Web Urls (Google Sheets) 3. Remote Search 4. Paste & Go
In this portfolio we'll cover all of them. Start by navigating to the Datasets menu and selecting New Dataset on the right side of the page. Provide a topic & description to get started.
Local Files Open the Dataset Wizard and select one or more files from the file selector. Alternatively you can drag & drop files right onto the panel.
Web Links Select the Web Links option to provide links to the wizard. You can add multiple links related to your topic.
Remote Search
Use our Remote Search feature to leverage the power of Claude, Gemini and Grok combined to find datasets relevant to your topic.
Select Results After the downloads complete select the web links or remote dataset results you want to import.
Download Selection This wizard page provides real-time status updates of your download. You can proceed with the wizard when they're all completed.
Define Columns A critical step in setting up datasets for correlations. Select columns to include in your dataset that you want available for drill-down in the Discovery details. By default all columns are selected but only numeric columns are used in correlations.
When setting up columns it's important to define the Join Key. This is the field in your dataset that will join with other datasets. Commonly time-series values are used, but you may also join by shared key such as stock market symbols.
Multiple join keys are supported for datasets whose primary keys are not unique, up to 4 compound primary keys can be checked.
Tip: When selecting a Year field of type Number, change the data type to Date to match it with other time-series datasets.
Paste & Go
Select the 'Skip the line' checkbox on the 1st page of the Dataset Wizard, then paste in your CSV or Excel clipboard data. Click 'Next' to skip directly to the Column Editor.
Datasets Summary This page summarizes your dataset ingest plan and provides an estimate of the cost in tokens. Select Create Datasets to complete the wizard and ingest your files.
Dataset Overview
After completing the dataset wizard your selected files will be ingested. Watch for their status to update in the My Datasets page.
Experiments Overview
From the main menu, click Experiments. This will show you an expandable overview of all of your Experiments. To create a new one, click New Experiment.
New Experiment
Using the Create Experiment workflow, users combine Datasets together and prepare to analyze each column against every other. Experiments can consist of Self correlations, which examine every column in a single Dataset against every other, Cross correlations which examine every column of every Dataset in the experiment, or Star correlations which examines an anchor dataset against every other. A Full correlations option is provided to merge all three correlation types.
Here we’re creating a Star Experiment for the S&P 500 Daily Time Series, so the Row Matching mechanism is Time Series. Other values include Shared Key, useful for connecting datasets with information like stock ticker symbols, and Row Offsets, used when the Datasets are from the same export process and known to be linked by cardinal offset.
Additional parameters to the Experiment workflow include Sample Size, which can be used to select idempotent samples of the target Datasets, Minimum Paired Rows and Minimum Sample Coverage which ensure healthy statistical assessments.
Experiments Overview
After the Create Experiments workflow is complete the user is presented with an overview of all of their Experiments. There may be many, as in this example of 50. Each Experiment row can be expanded to reveal the Discoveries it contains, and each Discovery row can be expanded to reveal its chart.
Experiment Details
Depending on the parameters of the Experiment, the columns in the Datasets and the number of rows connected by there may be anywhere from one Discovery to thousands. The Experiments detail page includes all the metadata and parameters of the Experiment, and a table of Discoveries. Click through to each one for a drill-down view.
Experiment Sparklines
Use the experiment sparklines view to see a collection of discovery scatterplots. Useful for selecting discoveries for drill-down.
Discovery Detail
Click-through on any Discovery to view a detailed page with metadata, scatterplots & line charts with drill-down capability and geospatial views where available. The datapoints, P-values, Granger Causality & other metadata are all part of the Discoveries that are analyzable.
Causality Analysis
Each Discovery can be analyzed with the Granger Analysis and AI Analysis buttons to explore causality.
