Tabular Data Anomaly-Detector Module

This module implements the rdk:component:sensor API in an anomaly-detector model. This repo can be used as a starter kit to implement your personal real time anomaly detection using a machine learning model trained with your data. This is an advanced topic and Viam basics such as configuring machines, components and services are expected to be known! If you are not familiar with these topics yet, I recommend to start with this beginner tutorial first and then come back.

Concept

The anomaly sensor takes an "actual" sensor as dependency/input and forwards it to the isolation forest onnx model for inference. The process is triggered by a client e.g. the Viam data manager polling the anomaly detector readings api:

Requirements

If you haven't done so, I recommend to familiarize with the topic of creating your own Viam resources. You can find the related documentation here: Create Your Own Modules.

To create your own development setup follow these steps (currently tested on Darwin/arm64):

1. Install viam-server on your machine Install Guide
2. Clone this repository to your machine or just download the sample.onnx file
3. In app.viam.com navigate to the Configuration tab of your machine
4. Add the ONNX module under add Service to your machine
5. Change the name to mlmodel in the prompt
6. In the automatically created mlmodel component enter the path to the previously downloaded sample.onnx file
7. Add a fake sensor
8. Add the anomaly detector module under add Component from the Viam registry
9. Add this configuration to the newly added attributes of the anomaly detector sensor:

{
  "features": [
    {
      "key": "a"
    },
    {
      "key": "Hour"
    },
    {
      "key": "Week_Day"
    },
    {
      "key": "Month_Day"
    },
    {
      "key": "Month"
    },
    {
      "key": "c",
      "rolling_mean": 3
    },
    {
      "key": "c",
      "lag": 1
    }
  ],
  "sensor": "fake-sensor",
  "model": "mlmodel"
}

6. Save the configuration and start the machine

Configure your anomaly-detector sensor

Attributes

The following attributes are available:

Name	Type	Required?	Description
sensor	string	Required	The sensor you want to apply the ml model to
model	string	Required	The ml model to be used
features	[struct]	Required	List of features
feature.key	string	Required	The key to the reading of your input/source sensor or one of "Year", "Month", "Month_Day", "Week_Day", "Hour", "Minute"
feature.rolling_mean	int	Optional	Number of readings to calculate the rolling mean of this feature
feature.lag	int	Optional	Adds the reading current-n.

Important

The order of features is consistent with the order used for passing them into the ML model. Make sure this maps to the order you have trained the model with!

Next steps

To train your own model with your data we have created a sample training script which should be easy to tune towards your particular data set: Model Training

Let us know if you are interested in this topic, have questions, struggle with setting it up, enhancing it etc.. We are happy to help!

Troubleshooting

N/A