Building a Model
The flow navigator is the heart of Akkio’s no-code AI. You can access the Flow Navigator by clicking on any flow.
In Akkio, a flow is an end-to-end AI model from data input to model deployment. The flow navigator is the visual interface used to connect data, build an AI model, and deploy, entirely without code.
Pro-Tip: If you want to use the same flow for different purposes, you can duplicate it from the “All Flows” homepage!
Lets start from scratch and create a new flow. Select the blue plus sign labeled 'Create New Flow'.
The first step in the process is to select what data you want to use to train your model. Machine Learning is based on a system of training and response, in this case we will need to feed the model reference data to teach it what an accurate set of outcomes looks like. If you are following this guide step by step you will already have setup your data in Selecting Model Data. The types of sources are noted here with links to setting up the relevant integrations when needed.
For this example we are going to use a standard table 'Telco-Customer-Churn.csv'. We will also assume that any data prep or cleanup has already been performed. This will then take us to the next stop: training the model.
You can build an AI model to make predictions on your connected data by clicking on the blue plus sign in the left-hand column.
This will take you to a selection page with the following options:
- Forecast (Beta)
- Detect Anomalies (Beta)
Then, you’ll need to select the column to predict.
For example, in the Telco-Customer-Churn dataset, that column is simply called “Churn.” This is an example of classification, as the model will predict that a customer will either churn, or not.
The “Training Mode” section gives you several options for model training:
- High quality
- Higher quality
The quality options are ordered in time taken to train. This won’t necessarily lead to higher accuracy, due to a phenomenon known as “overfitting,” which is when a machine learning model only learns the training data well, but does not generalize to new data.
You will also note advanced features below the Training Mode selector, these are detailed on a separate page.
Once you select a training quality the model will automatically complete its training. The output should look like this:
If the data doesn't look right or you don't like the results it worth going back into the data set and taking a look at your source material. Chat Data Prep can help you clean up your data for better results as needed. Then just come back to here and retrain your model to see the improvement.
Once your model is trained and you are happy with it you can setup your deployment. Select the blue plus sign to return to the segment select screen and pick from the available 'Deploy Flow' Options.
- Web App
- Google Sheets
- Google BigQuery
Below you can see the results from the demo when displayed on the Web App. The Akkio generated Web App allows for single or bulk input of new data to the model. Now you will be able to predict churn easily based on the training just completed.
Datasets are the fundamental building blocks of Akkio, and indeed fundamental to all machine learning tasks.
Akkio comes with several datasets pre-included as part of “demo” flows, so you can see how the product works. This includes:
- An insurance dataset
- An employee attrition dataset
- A telecom customer churn dataset
- A credit card dataset
- A historic conversions dataset
- A restaurant reviews dataset
- An avocado prices dataset
You can easily add or delete datasets, whether from a CSV, Excel file, JSON file, Google Sheets, Google Big Query, Salesforce, Snowflake, or Hubspot.