Hi, The August meetup (8/25/2019)
Vijay Kumar gave a quick intro about ChennaiPy group and its motives. *1.Topic : Central service registry to enable service discovery in Microservices architecture with Django & PostgreSQL* Sandhya balakrishnan from Mad Street Den gave a talk on Central service registry .She shared about the problems and use cases they faced , how they overcame them. She started with Micro services vs Monolithic comparison in layman terms using Facebook components example and how Micro service architecture helps in independent deployment , auto scale and flexibility in tech stack .But the challenge lies in the intercommunication among different components ;which they have solved using central registry .She also compared it with registration desk & load balancers for better understandability. vue.ai is for retail sites : using AI to auto generate and recommend products. They use Vuetag components for Feature extraction, input parsing and Vuemodel component for auto generation . They have overcome another main challenge of notification system using publish-subscribe model; where the central registry passes the updates to all micro services. Also AWS EC2 instances have helped them for load balancing.They are now trying on moving to Dockers. Questions were raised on what if central registry fails? and How development teams communicate with each other ? and also on security mechanisms are carried out. *2. Topic:POWER OF ENSEMBLES* Padma Priya from PayPal started her talk describing ML timeline and evolution of ML starting from Baye's theorem in 1763, Turing test to Hadoop and Kaggle. She used the Elephant and Blind people analogy to explain the ensemble concept, how different perspectives are combined together to present more accurate answers. In the current real-world use cases with huge, non-linear and different types of more complex data , traditional ML models do not apply well. Hence ensemble techniques are used to have low bias, low variance and production ready models. The concept is to build multiple diverse meta models and finally aggregate them . The meta models can be diversified by either feeding different subsets of data to model or different models on the same data or even with different hyper parameters on same model. Hence this also ensures more generalization. She also explained about Bagging, boosting and stacking. Bagging is a parallel learning model with bootstrapping and aggregating. Boosting is a sequential learning model. Stacking and blending splits and trains in two folds. She also stressed that domain knowledge is very much required in noise handling, new data and new features coming in . *3. Topic:* *MQTT* Ajay kumar explained about a machine to machine protocol MQTT with publish/subscribe system which can be majorily benefitted in IoT systems. The MQTT (Message Queuing Telemetry Transport) also ensures assured delivery through acknowledgement. Ajay demonstrated open source Eclipse Mosquito and paho-mqtt libraries with real-time demo on sampled data on four components with publish/subscribe model. A second demo on how messages to different subscribers were updated in the dashboard, without the page getting refreshed. He also showed a live demo for quick group chat , involving the audience to message in the group chat URL powered by MQTT protocol and explained how Facebook messenger and other companies like Amazon AWS and Azure are utilizing this protocol. The session was followed by discussions and comparison with other existing protocols and Apache Kafka by the audience. *PyCon* Vijay Kumar announced about the upcoming PyCon India conference and explained about tech talks and dev sprints. Volunteers are invited for the event support as well . Thanks, Chandnika R
_______________________________________________ Chennaipy mailing list Chennaipy@python.org https://mail.python.org/mailman/listinfo/chennaipy
