{"id":597,"date":"2020-05-29T05:40:47","date_gmt":"2020-05-29T05:40:47","guid":{"rendered":"https:\/\/datascientists.info\/?p=597"},"modified":"2020-05-29T05:40:47","modified_gmt":"2020-05-29T05:40:47","slug":"bringing-machine-learning-models-into-production","status":"publish","type":"post","link":"https:\/\/datascientists.info\/index.php\/2020\/05\/29\/bringing-machine-learning-models-into-production\/","title":{"rendered":"Bringing machine learning models into production"},"content":{"rendered":"\n

Developing and bringing machine learning models into production is a task with a lot of challenges. These include model and attribute selection, dealing with missing values, normalization and others.<\/p>\n\n\n\n

Finding a workflow that puts all the gears, from data preprocessing and analysis over building models and selecting the best performing one to serving the model in a real time API, into motion is the one I want to discuss here.<\/p>\n\n\n\n

Life cycle of a machine learning model<\/h2>\n\n\n\n

The life cycle of machine learning is basically described by the iteration of the following four steps.<\/p>\n\n\n\n

\"Life
Machine learning steps<\/figcaption><\/figure><\/div>\n\n\n\n

Each of these steps is under constant evaluation. Especially in case model performance can be enhanced by adding different data attributes or preprocessing methods.<\/p>\n\n\n\n

For the presented approach we split the process of modeling into two other parts. Part one contains the above mentioned four steps and we call it Manual Run Modeling. Step two is automating the steps of part one.<\/p>\n\n\n\n

\"machine<\/figure>\n\n\n\n

Manual Run Modeling<\/h2>\n\n\n\n

In the manual part we start by analysing our new task. After that we come up with a hypothesis we want to prove and test.<\/p>\n\n\n\n

Development and Prototyping Environment<\/h3>\n\n\n\n

First we set up a development environment for working on the new task. For this we spin up a Jupyter<\/a> notebook server. We deploy it on Google Cloud AI Platform. It provides ready to run containers for Jupyter. The notebook approach enables us to develop fast and share results with the team using a browser. With the ability to easily visualize data inline in a notebook, this approach is especially useful in the data extraction and preprocessing process.<\/p>\n\n\n\n

Data preparation and visualization<\/h4>\n\n\n\n

Python provides some nice packages for generating graphics on data for faster insights. This speeds up our prototyping process in the notebook. We are especially fond of using Seaborn<\/a>.<\/p>\n\n\n\n

We load the data identified for this model into a dataframe in the notebook. After that we start looking at each attribute and its values, often in combination with the other attributes. For a first overview we use a pairplot provided by Seaborn.<\/p>\n\n\n\n

We use a combination of visualizations, such as e.g. a correlation matrix. Then we decide which attributes to use and how to handle outliers and missing values. Finally we use one hot encoding for categorical attributes and normalize the continuous attributes to create the input into our models.<\/p>\n\n\n\n

\"pairplot<\/figure>\n\n\n\n

Model Selection and Evaluation<\/h4>\n\n\n\n

When the data is ready we choose several models to find a solution for our problem. These models can range from a multilinear regression model over random forests to deep neural networks with Tensorflow.<\/p>\n\n\n\n

After splitting the data for training, evaluation and test we decide on a measure each model has to optimize for, e.g. mean squared error or precision. This depends on the kind of problem. Once we identified the best performing model we start by transforming the code for Google Cloud AI Platform.<\/p>\n\n\n\n

RT Prediction Deployment<\/h2>\n\n\n\n

After manual evaluation of preprocessing and modeling, we start the task of automating training and deployment for bringing machine learning models into production. This can be split into three tasks:<\/p>\n\n\n\n