Learnings

Deployment Deployment is a process of running an application on a server or pipeline Given a model object, how to make it part of your application. It consist of components: Deploy model Low level solution Off the shelf solution (High level solution) Serving model Batch serving Low latency serving Batch Serving Pros: Very flexible Easily set up, you can put the model object/data anywhere Cons: High latency Process: Read input Run inference Return output Low Latency Serving Pros: Low latency Cons: Rigid schema Infrastructure setup overhead Only handles relatively simple model? Complexity: Communication interface setup Format (grpc/json) Security Multiple model deployment management Hybrid Still have the same complexity as low latency but with More accurate model Feasible infrastructure Input doesn’t need to contain all features Packages To properly productionise deploy and serving model, we may want to consider requirements related to Model versioning Model metadata Model artifact Lo...

https://arxiv.org/abs/1908.07442 With the growth of the neural networks, DNN is showing a good performance in text, image, audio data. However, for the everyday tabular data, we've yet to see the success compare to XGB LGBM. Tree based models have advantages in: decision manifolds, like a super hyperplane (infinite expandable, cuts the table well) interpretable fast to train DNN able to encode data (representation learning) reduce feature engineering online learning discuss: reasons why DNN not performant on tabular data non-linear, not restriction on convergence, easy overfit compare to tree ensemble adding more layers can cause overparameterization, this may be why it isn't performant in tabular data. How good would it be if we can have a framework that's both end-to-end, representation and can do online update for tabular data build decision boundary using DNN tree:  DNN: we can treat the mask as a decision boundary the mask+FC+ReLU is like a vanilla decision tree this i...