Perform research and design of algorithmic models
About
This unit is about performing research and designing a variety of algorithmic models for internal and external clients
Scope
define hypothesis, select model, prototype and design
Define hypothesis
- identify the objective of the analysis
- develop a hypothesis based on the objective of the analysis
- identify suitable libraries, packages, frameworks, applications to address the objective
Select model
- identify mode of learning, i.e. supervised or unsupervised
- conduct research on existing statistical models to evaluate fitment with the objective
- depending on the use case, identify if neural networks or deep learning models can be built
- optimize the existing statistical models as per need
- identify suitable statistical models on the basis of data volumes and key variables
- define connectors or combinations of key variables for each statistical model
Prototype and design
- determine and collect the training data
- design and prototype algorithmic model
- identify and resolve overfitting or underfitting of algorithmic model
- identify and resolve residual and dispersion errors with data
- define data flows such as human-in-the-loop constraints required to reinforce algorithmic models
- define and quantify success metrics for the algorithmic model
- create documentation on designed algorithmic models for future references and versioning
- retrain datasets that have been used for supervised learning on a continuous basis
- validate designed models using appropriate tools and processes
- Iterate the process to fine-tune the model til the desired quality of output or performance is achieved
Required Knowledge & Understanding
Technical Skills
- how to develop experimental and analytical plans for data modeling
- the use of strong baselines
- how to accurately determine cause and effect relations
- different probability theory concepts such as probability distributions, statistical significance, hypothesis testing and regression
- different Bayesian thinking concepts such as conditional probability, priors and posteriors, and maximum likelihood
- strong research experience in deep learning, reinforcement learning and other machine learning algorithms and their usage
- different programming languages that can be used to design algorithmic models such as python, ruby, C, java, c++ or c#
- different use cases and the suitability of various algorithmic models to address them
- how to build and test a hypothesis
- when to use supervised or unsupervised learning
- how to evaluate data volumes and key variables
- how to define combinations of key variables
- how to optimize overfitting or underfitting of algorithmic models
- how to optimize residual and dispersion errors in algorithmic models
- how to define data flows such as human-in-the-loop constraints required to reinforce algorithmic models
- different cloud or distributed computing platforms such as AWS, Azure, Hadoop, their affiliated services and how to use these
- how to identify and refer anomalies in data
- how to work on various operating systems such as linux, ubuntu, or windows
Soft Skills
Analytical Thinking
impact analysis of the various actions performed and disseminating relevant information to others. analyze data, models and understand its implications on business performance
Attention to Detail
check your work is complete and free from errors