maadstmlmedia

Multi-Agent Accelerator for Data Science (MAADS): Transactional Machine Learning


Keywords
multi-agent, transactional, machine, learning, audio, video, images, data, streams, science, optimization, prescriptive, analytics, automl, auto-ml, artificial, intelligence, predictive, advanced
License
MIT
Install
pip install maadstmlmedia==1.2

Documentation

Transactional Machine Learning : The Machine Learning Platform for Data Streams

Transactional Machine Learning (TML) using Data Streams and AutoML is a platform for building and streaming cloud native solutions using Apache Kafka or Redpanda as the data backbone, with Kubernetes and Docker as core infrastucture components, running on Confluent, AWS, GCP, AZURE, for advanced machine learning solutions using transactional data to learn from, and provide insights, quickly and continuously to any number of devices and humans in any format!

TML Is Based On the Belief that "Fast data requires fast machine learning for fast decision-making". TML gives rise in the industy to a Data Stream Scientist versus a Data Scientist in conventional machine learning (CML).

TML Book Details Found on Publisher's site

TML Video: Youtube

Apply data preprocessing and auto machine learning to data streams and create transactional machine learning (TML) solutions that are:

1. frictionless: require minimal to no human intervention

2. elastic: machine learning solutions that can scale up or down using Kubernetes to control or enhance the number of data streams, algorithms (or machine learning models) and predictions instantly and continuously.

TML is ideal when data are highly erratic (nonlinear) and you want the machine to learn from the latest dataset by creating sliding windows of training datasets and auto creating micro-machine learning models quickly, that can be easily scaled, managed and the insights used immediately from any device! There are many TML use cases such as:

1. IoT: Capture real-time, fast, data, and build custom micro-machine learning models for every IoT device specific to the environment that the device operates in and predict failures, optimize device settings, and more.

2. IoT Edge: TML is ideal for edge devices with No Internet connections. Simply use the On-Prem version of TML software, with On-Prem Kafka and create large and powerful, real-time, edge solutions.

3. HealthCare: TML is ideal for health data processing for patients, payers, and providers. Open access to health data has been mandated by CMS, which opens up enormous opportunities for TML.

4. Banking/Finance Fraud Detect: Detect fraud using unsupervised learning on data streams and process millions of transactions for fraud - see the LinkedIn blog

5. Financial Trading: Use TML to analyse stock prices and predict sub-second stock prices!

6. Pricing: Use TML to build optimal pricing of products at scale.

7. Oil/Gas: Use TML to optimize oil drilling operations sub-second and drill oil wells faster and cheaper with minimal downtime

8. SO MUCH MORE...

The above usecases are not possible with conventional machine learning methods that require frequent human interventions that create lots of friction, and not very elastic.

By using Apache Kafka On-Premise many advanced, and large, TML usecases are 80-90% cheaper than cloud-native usecases, mainly because storage, compute, Egress/Ingress and Kafka partitions are localized. Given Compute and Storage are extremely low-cost On-Premise solutions with TML are on the rise. TML On-Prem is ideal for small companies or startups that do not want to incur large cloud costs but still want to provide TML solutions to their customers.

Strengthen your knowledge of the inner workings of TML solutions using data streams with auto machine learning integrated with Apache Kafka. You will be at the forefront of an exciting area of machine learning that is focused on speed of data and algorithm creation, scale, and automation.


Create Your First TML Solution with Kafka by Downloading the Technologies Below

WATCH The TML Instructional Video: Setup, Configuration, Execution, Visualization

  1. MAADS-VIPER: https://www.confluent.io/hub/oticsinc/maads-viper (Official Kafka connector for TML - Linux version). Latest Linux/Windows/MAC version can be found above.). More information here.
  2. MAADS-VIPERviz: Streaming Visualization for Windows/Linux/MAC versions
  3. MAADS-HPDE: AutoML for Windows/Linux/MAC versions available
  4. MAADS-Python Library: https://pypi.org/project/maadstml/ (NOTE: You need Python IDE installed: tested with Python up to v.3.8)
  5. Create a Kafka Cluster at Confluent Cloud: https://www.confluent.io/confluent-cloud
  6. Users can also, directly, use MAADS-VIPER and Kafka services on Amazon AWS, Microsoft Azure, and Google (GCP)

Contact:

For any help and additional information, or if your token has expired you can e-mail: info@otics.ca or goto http://www.otics.ca, and we would be happy to help you! OTICS will provide 1 hour free developer session on TML if needed.


Watch University of Calgary Lecture on TML to Software Engineering Graduate Students

Read Confluent blog

Read Medium blog

Watch ApacheCon Presentation on TML - Apache Software Foundation

Read Fast Big Data Visualization blog


IF CREATING PRODUCTION TML SOLUTIONS: SEE HERE FOR CODE EXAMPLES


EXAMPLE TML PYTHON CODE: You can literally build extremely powerful, distributed, and scalable cloud-based machine learning solutions with the code below for your business use case of any size with low-code and low-cost!

CODE SET 1: This set of programs will go through an example of predicting and optimizing Foot Traffic at ~11,000 Walmart Stores.

Step 1:

Produce Walmart Data to Kafka Cluster (Let this run for 5 minutes or so THEN run the Machine Learning code next)

Step 2:

Walmart Foot Traffic TML (Let this run for 5 minutes or so THEN run the Prediction/Optimization code next)

Step 3:

Perform Walmart Foot Traffic Prediction and Optimization

Step 4:

To Visualize the results in Step 3 you need to run MAADS-VIPER Visualization (VIPERviz) and then enter the following URL For:

Visualize Predictions: https://127.0.0.1:8003/prediction.html?topic=otics-tmlbook-walmartretail-foottrafic-prediction-results-output&offset=-1&groupid=&rollbackoffset=10&topictype=prediction&append=0&secure=1&consumerid=[Enter Consumer ID for Topic=otics-tmlbook-walmartretail-foottrafic-prediction-results-output]&vipertoken=hivmg1TMR1zS1ZHVqF4s83Zq1rDtsZKh9pEULHnLR0BXPlaPEMZBEAyC7TY0

Visualize Optimization: https://127.0.0.1:8003/optimization.html?topic=otics-tmlbook-walmartretail-foottrafic-optimization-results-output&offset=-1&groupid=&rollbackoffset=10&topictype=optimization&secure=1&append=0&consumerid=[Enter Consumer ID for Topic=otics-tmlbook-walmartretail-foottrafic-prediction-results-output]&vipertoken=hivmg1TMR1zS1ZHVqF4s83Zq1rDtsZKh9pEULHnLR0BXPlaPEMZBEAyC7TY0

The Above Assumes:

  1. You have created a Kafka cluster in Confluent Cloud (Or AWS, Microsoft or Google Cloud)
  2. You have MAADSViz running on IP: 127.0.0.1 and listening on Port: 8003
  3. You downloaded views zip and extracted contents to viperviz/views folder
  4. You added the consumer id for Topic=otics-tmlbook-walmartretail-foottrafic-prediction-results-output and Topic=otics-tmlbook-walmartretail-foottrafic-optimization-results-output
  5. This Consumer IDs are printed out for you in the Python Program in Step 1 c)

CODE SET 2: This set of program will perform Bank Fraud detection in 50 Bank accounts and 5 fields in each transaction. It will detect fraud in real-time.

Step 1:

Produce Bank Account Data to Kafka Cluster (Let this run for 5 minutes or so THEN run the Anomaly Detection code next)

Step 2:

Perform Transactional Bank Fraud Detection on Streaming Data This will use multi-threading in Python

Step 3:

Visualize Anomalies:

https://127.0.0.1:8003/anomaly.html?topic=otics-tmlbook-anomalydataresults&offset=-1&rollbackoffset=20&append=0&topictype=anomaly&secure=1&groupid=&consumerid=[Enter your Consumer ID for otics-tmlbook-anomalydataresults]&vipertoken=hivmg1TMR1zS1ZHVqF4s83Zq1rDtsZKh9pEULHnLR0BXPlaPEMZBEAyC7TY0

The Above Assumes:

  1. You have created a Kafka cluster in Confluent Cloud (Or AWS, Microsoft or Google Cloud)
  2. You have MAADSViz running on IP: 127.0.0.1 and listening on Port: 8003
  3. You downloaded views zip and extracted contents to viperviz/views folder
  4. You added the consumer id for Topic=otics-tmlbook-anomalydataresults
  5. This Consumer IDs are printed out for you in the Python Program in Step 2 b)

NOTE: Please monitor your Cloud Billing/Payments - DELETE YOUR CLUSTER WHEN YOU ARE DONE. DO NOT LET YOUR CLUSTER RUN IF YOU ARE NOT USING IT. The above programs will auto create all data very quickly. So you can DELETE your cluster immediately. Confluent will give you $200 free cloud credits. The above programs will consume a low fraction of this free $$.