{"id":2167,"date":"2019-12-31T02:23:16","date_gmt":"2019-12-31T02:23:16","guid":{"rendered":"http:\/\/kusuaks7\/?p=1772"},"modified":"2024-02-01T03:20:45","modified_gmt":"2024-02-01T03:20:45","slug":"take-your-machine-learning-models-to-production-with-these-5-simple-steps","status":"publish","type":"post","link":"https:\/\/www.experfy.com\/blog\/ai-ml\/take-your-machine-learning-models-to-production-with-these-5-simple-steps\/","title":{"rendered":"Take your Machine Learning Models to Production with these 5 simple steps"},"content":{"rendered":"\t\t
\n\t\t\t\t\t\t
\n\t\t\t\t\t\t
\n\t\t\t\t\t
\n\t\t\t
\n\t\t\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t\t\t\t\t\t
Creating a great machine learning system is an art.<\/blockquote>\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t\t\t\t\t\tThere are a lot of things to consider while building a great machine learning system. But often it happens that we as data scientists only worry about certain parts of the project.\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t\t\t\t\t\tBut do we ever think about how we will deploy our models once we have them?<\/em><\/strong>\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t\t\t\t\t\tI have seen a lot of ML projects, and a lot of them are doomed to fail as they don\u2019t have a set plan for production from the onset.\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t\t\t\t\t\tThis post is about the process requirements for a successful ML project \u2014 One that goes to production.\n\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t\t

\n

1. Establish a Baseline at the onset<\/h2><\/h2>\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t\t\t\t\t\tYou don\u2019t really have to have a model to get the baseline results.\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t\t\t\t\t\tLet us say we will be using RMSE as an evaluation metric for our time series models. We evaluated the model on the test set, and the RMSE came out to be 3.64.\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t\t\t\t\t\tIs 3.64 a good RMSE? How do we know? We need a baseline RMSE.\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t\t\t\t\t\tThis could come from a currently employed model for the same task. Or by using some very simple heuristic. For a Time series model, a baseline to defeat is last day prediction. i.e., predict the number on the previous day.\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t\t\t\t\t\tOr how about Image classification task. Take 1000 labelled samples and get them classified by humans. And Human accuracy can be your Baseline. If a human is not able to get a 70% prediction accuracy on the task, you can always think of automating a process if your models reach a similar level.\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t\t\t\t\t\tLearning<\/em><\/strong>: Try to be aware of the results you are going to get before you create your models. Setting some out of the world expectations is only going to disappoint you and your client.\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t\t

2. Continuous Integration is the Way Forward<\/h2><\/h2>\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t\"\"\t\t\t\t\t\t\t\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t\t\t\t\t\t\nYou have created your model now. It performs better than the baseline\/your current model on your local test dataset. Should we go forward?\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t
\n\t\t\t\t
\n\t\t\t\t\t\t\t\t\tWe have two choices-\n
    \n \t
  1. Go into an endless loop in improving our model further.<\/li>\n \t
  2. Test our model in production settings, get more insights about what could go wrong and then continue improving our model with\u00a0continuous integration.<\/em><\/strong><\/li>\n<\/ol>\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t
    \n\t\t\t\t
    \n\t\t\t\t\t\t\t\t\tI am a fan of the second approach. In his awesome\u00a0third course<\/a>\u00a0named Structuring Machine learning projects in the Coursera\u00a0Deep Learning Specialization<\/a>, Andrew Ng says \u2014\n
    \u201cDon\u2019t start off trying to design and build the perfect system. Instead, build and train a basic system quickly \u2014 perhaps in just a few days. Even if the basic system is far from the \u201cbest\u201d system you can build, it is valuable to examine how the basic system functions: you will quickly find clues that show you the most promising directions in which to invest your time.\u201d<\/blockquote>\nDone is better than perfect.\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t
    \n\t\t\t\t
    \n\t\t\t\t\t\t\t\t\tLearning<\/em><\/strong>: If your new model is better than the current model in production or your new model is better than the baseline, it doesn\u2019t make sense to wait to go to production.\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t
    \n\t\t\t\t
    \n\t\t\t\t\t

    3. Your model might break into Production<\/h2><\/h2>\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t
    \n\t\t\t\t
    \n\t\t\t\t\t\t\t\t\tIs your model better than the Baseline? It performed better on the local test dataset, but will it really work well on the whole?\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t
    \n\t\t\t\t
    \n\t\t\t\t\t\t\t\t\tTo test the validity of your assumption that your model is better than the existing model, you can set up an\u00a0A\/B test<\/em><\/strong>. Some users(Test group)see predictions from your model while some users(Control) see the predictions from the previous model.\n\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t
    \n\t\t\t\t
    \n\t\t\t\t\t\t\t\t\tIn fact, this is the right way to deploy your model. And you might find that indeed your model is not as good as it seems.\n
    Being wrong is not wrong really, what\u2019s wrong is to not anticipate that we could be wrong.<\/blockquote>\t\t\t\t\t\t\t\t<\/div>\n\t\t\t\t<\/div>\n\t\t\t\t
    \n\t\t\t\t
    \n\t\t\t\t\t\t\t\t\tIt is hard to point out the real reason behind why your model performs poorly in production settings, but some causes could be:\n