Real world examples of Machine Learning? [closed]

Question

Recently i was doing the Machine Learning course at Coursera by Prof. Andrew Ng. After doing this course i have understand the basics of Machine Learning Algorithms, but i have the following questions:

• Where can i find the Real world Machine Learning use case examples?

• What tools or framework are used in Industry/Production for Machine
  Learning projects?

• How Machine Learning models are used or deploy in production?

• How to become Data Scientist? Or What should i do next?

Any suggestion,books,courses or tutorial links will be highly appreciated.

Answer 1

Congrats for completing the Machine Learning course by Andrew Ng, longtime back I have also done this awesome course. Anyway, so I will answer your question one by one, although there are few questions that are interrelated.

Q-1) Where can I find the Real world Machine Learning use case examples?

Here are a few links where you can find tutorials of machine learning with real-world use cases:

• Machine Learning example Scikit learn: http://scikit-learn.org/stable/tutorial/basic/tutorial.html

• Machine Learning tutorials H20: https://github.com/h2oai/h2o-tutorials

• Sagemaker Machine Learning & Deep Learning example: https://github.com/awslabs/amazon-sagemaker-examples

• Tutorials on real-world Machine Learning Case Studies

Q-2) What tools or framework are used in Industry/Production for Machine Learning projects?

There are a wide variety of tools or framework that are used in Industry level like:

Machine Learning:

• R (mostly used in academia nowadays)
• Python(Sci-kit learn)
• GraphLab
• Apache Mahout
• Spark MLlib
• H20

Deep Learning:

• Tensorflow and Keras
• Mxnet
• Pytorch
• DeepLearning4j
• Analytics Zoo (mainly for deep learning with big data using spark)

While R,Scikit learn,GraphLab works great on single machine and most popular choices among data scientist or machine learning practitioners, but Mahout, H20 and recently Spark (MLlib) has gained a lot of popularity in this era of Big Data, where you want to do machine learning on large dataset that will not fit on single machine.

Also, there are some other tools like Weka, Rapid Miner for GUI based workflow of machine learning work.

Choice of this tool or framework really depends on the factors like project requirement, team members knowledge about the tool/language, also ease of development and scalability of deployment.

Q-3) How Machine Learning models are used or deploy in production?

In production, you have to first build a model, validate & evaluate that model, then the model is most finally deployed as web/rest service to be used by other applications/services. Deploying a machine learning model depends on a lot of factors such as-

• Is the model trained offline? Or are you deploying an online learning model?
• How often you will retrain your model?
• How would you test your newer version of the model? - A/B testing or Bandit variation.
• Along with other generic things - latency, throughput, data input/output format etc.

There are some cloud-based machine learning service provider like Azure ML(https://studio.azureml.net/) BigML(https://bigml.com/) etc, where you can upload your dataset,do some data processing, train|validate|evaluate your machine learning model and then finally deploy it as web service in the cloud.

Also all major cloud platform (aws, google cloud, azure) nowadays provide you with a machine learning platform, where you can build your own model, evaluate them and then finally deploy it in the cloud. It gives you the flexibility to build the model with almost all major machine learning or deep learning frameworks, and as per your requirement gives you the flexibility to deploy ( what type of server/containers, number of inference/prediction server, etc).

Amazon SageMaker:

[+] https://aws.amazon.com/sagemaker/

Google Cloud Machine Learning (ML) Engine:

[+] https://cloud.google.com/ml-engine/

Q-4) How to become Data Scientist? Or What should I do next?

It's a million dollar question and lots of google serach on this question..haha.. I will try to give you a short and concise answer based on my knowledge. First of all Data Science is much broader field of study,that comprise of the following common steps:

• Business understanding or Questioning Phase
• Data gathering or acquiring
• Data processing and preparation
• Model Building
• Validation and Evaluation

Along with this, you also need to do Model Retraining to depending upon the change of data variability, or you can deploy online learning model(which will adapt itself based on the data that it is seeing).

But basic ingredients for anyone to become data scientist/machine learning practitioner is to have the curiosity about data (i.e to understand the data & find valuable knowledge out of it). Neither there is a shortcut to becoming a data scientist nor there is any course that will make you become data scientist overnight.

There is no predefined role/scope what a data science person should be knowing or doing on a day to day basis in a company. Different industry or company has its own job requirement/description for a data scientist depending on their business problem.

A good versatile data scientist must have the following skills in order to sustain confidently across various industry and succeed in his/her career:

• Good knowledge in Statistics (including a little bit of Bayesian)- essential during EDA phase.

• Maths (especially linear algebra,matrix,vector, multivariate calculus): https://www.coursera.org/specializations/mathematics-machine-learning

• Good practical knowledge of machine learning algorithms: https://www.coursera.org/specializations/machine-learning

• Some deep learning & reinforcement learning knowledge. Stanford Deep Learning course: http://cs231n.stanford.edu/ and youtube video of this course by Andrej Karpathy- https://www.youtube.com/watch?v=vT1JzLTH4G4&list=PLC1qU-LWwrF64f4QKQT-Vg5Wr4qEE1Zxk . Also DeepLearning Coursera: https://www.coursera.org/specializations/deep-learning and Reinforcement Learning course- Berkley University: http://rail.eecs.berkeley.edu/deeprlcourse/

• Large dataset analysis through big data tools like Spark and SQL. Machine Learning with Big Data: https://www.coursera.org/learn/machine-learning-applications-big-data

• Curious mind to explore data and learn new things (to stay up to date with the latest innovation in this domain).

• And some Business domain knowledge- good to have (Optional)

---

The best way is to play with data or do some real-world projects. Lots of real-world datasets available publicly, you can pick a dataset of your choice of interest. Also, you can test your skill and expertise by participating in Machine Learning and Data Science competition at Kaggle.

To gain some knowledge about data science, how it works along with some hand on exercise, you can try course online like:

https://www.edx.org/course/introduction-computational-thinking-data-mitx-6-00-2x-2

https://www.edx.org/course/data-science-machine-learning-essentials-microsoft-dat203x