Hello there,
I am currently working on a project that involves implementing Hierarchical Temporal Memory for real time anomaly detection in an Industrial Internet of Things system. The goal is to monitor sensor data from various industrial machines and detect anomalies that could indicate potential failures or inefficiencies.
What are the best practices for preparing sensor data for HTM? 
each with different sampling rates. How should I handle the different scales and frequencies of the data?
Given that HTM is an unsupervised learning model; how should I approach the initial training phase?
Is there a specific amount of historical data that I should feed into the system to ensure accurate anomaly detection from the start?
I am planning to integrate HTM with a real time streaming platform like Apache Kafka or MQTT. Are there any known challenges or performance bottlenecks when using HTM for real time processing? How can I optimize HTMs performance in such a scenario?
Also; I have gone through this post; https://discourse.numenta.org/t/looking-for-guidance-on-applying-hierarchical-temporal-memory-htm-to-iot-data-anomaly-detection-mlops/ which definitely helped me out a lot.
Once anomalies are detected; how do you recommend setting the thresholds for what constitutes a significant anomaly? Is there a standard approach for fine-tuning these thresholds in an industrial setting?
Thank you in advance for your help and assistance. 