HPDE performs AutoML on (training) data streams. To start HPDE run the command (on MacOS, Windows or Linux environments):
- Command: [HPDE executable] [host] [port]
- Create unlimited number of HPDE instances for massive scale - fully compatible with microservices architecture for load shedding
- HPDE can be accessed via MAADS python library or REST API
HPDE will run the following supervised algorithms:
- Neural networks
- Logistic regressions/Decision Trees
- Gradient boosting regressions
- Linear/Non-Linear regressions
- Mathematical Optimization - if you want to find Global Optimal values for the independent variables and do prescritive analytics
HPDE will also fine tune all of the hyperparameters.
HPDE will also run unsupervised algorithms on data streams. Unsupervised algorithms are especially important for data streams because it is not always possible to classify data for supervised learning. For example, if doing Fraud Detection using TML, then this would require past knowledge of fraud, but this is not always possible with real-time data streams. So HPDE performs advanced unsupervised learning using:
- Peer Group Analysis (PGA)
- Break Point Analysis (BPA)
There are enormous advantages to performing unsupervised learning on data streams such as:
- At Scale Fraud (anomaly) detection without classifying data (no knowledge of historical fraud is needed)
- Real-Time Fraud (anomaly) detection
- Distributed Fraud (anomaly) detection
Examples of TML for Fraud (anomaly) detection:
- Transactional Banking - as transactions are done in real-time, TML solutions using HPDE can detect Fraud (anomalies) is seconds
- IoT predictive asset maintainennce - detect possible equipment failure on millions of IoT devices in real-time
- Product Failure - detect possible product failure and increase product quality
Lot more use cases for both supervised and unsupervised learning using HPDE with VIPER on data streams.