Getting started with Infer
Learn how to use SQL-Inf and Infer to do explainable predictions with ease. If you are new to Infer, this is a good place to start.
- 4 months ago
- 4 min read
The first thing to do is to setup an account with Infer.
To do this simply head to https://app.getinfer.io and click the sign up button.
The first time you login we will ask you a few questions about your motivations and needs for using the app. Please fill these out 🙏 this really helps us make the app the best it can be for you.
Now you are logged in and ready to infer!
The first time you login you will see a set of demo projects - each of these illustrating different use cases and types of analysis.
Churn Analysis: This example illustrates how to do churn analysis on a customer data set using the PREDICT and EXPLAIN commands to both predict churn and understand what drives churn. It also shows how to aggregate and group outputs to estimate expected loss in the near term, based on the prediction of churn.
Customer Lifetime Value: Illustrates how to define a Customer Lifetime Value using an ecommerce data set and how to use it to predict and understand what drives the lifetime value of customers. It also shows how to use SHAP to build a more detailed picture of what drives the LTV for geographical segmentations.
Customer Feedback: Uses the text analysis components of SQL-inf, in particular the SENTIMENT command, to show how you can use SQL-inf to understand customer reviews. It also shows how to combine SENTIMENT with EXPLAIN and PREDICT to understand what customer behaviors drive certain sentiments
To get started with your own analysis the first thing you need to do is connect or upload your data. Currently the only method supported is to upload data as CSV files.
If you don't have a dataset at hand try this one. It is a nice dataset of red wine quality scores coming from this this paper.
Now it is time to run your first piece of analysis.
First, select your data set in the left dataset menu. Then go to the codebox in the middle and run an SQL-inf query.
If you used the above red wine data set then simply copy-pasting the following command into the codebox and click the "Run" button.
Volia! You have run your first SQL-inf command!
You can learn a lot more about this example and how to explain and better understand the prediction by going through the Prediction tutorial.
Once you have run your analysis you can view and explore the results.
Note that each SQL-inf command run is saved and can be accessed in the query bar next to the code-box. Here you can also pin, rename and delete previously analysis.
Below the codebox, you will see the results panel. The results panel displays a visualisation dependent on the type of analysis run - in the case of a prediction it will show the prediction plotted against an input column. You can plot different columns against the prediction by clicking on the column headers in the results table or by using the arrows in the top-right corner of the result panel.
Below the visualisation you will see the output of the analysis in table form. This is the raw output of the SQL-inf command.
Once you have iterated on your analysis and found something interesting you probably want to export it to another platform for further exploration or dashboarding
To do this simply click on the "Export Results" button above the results table. The result set will then be downloaded as a csv-file.
You can also click on the "Download Graph" button above the graph to download the visualisation as an image file.
We will soon offer the ability to connect your Infer account directly to your data warehouse, but for now you need to prepare your data externally via another tool and upload it as a single table, in the form of a csv.
Another way would be to use the template "Predict" from the template code-box dropdown. You can then select quality as the prediction field, set the model='reg' option and click "Run".
You can view the selected dataset before running the analysis by clicking the "Input Data" tab in the result section below the codebox.