Workflow of a Machine Learning project
Make a ML project could be overwhelming, because could be difficult to know where to start. But there are a serie of steps that can help you to make the process easier, and that serie of steps is the workflow of a ML project, ¡Let’s see it!
Before start working with a ML project, you need to know what is the goal you are trying to achieve. Remember that we can see a ML model as an abstraction of a mathematical function. In a mathematical function, we give it an input and returns an output. And is the same with an ML model, you give it input data and returns you output data. The problem you try to solve is the one that will designate which and what type the input and output values will be.
And also, you need to be sure that the project is feasible. Because there are wonderful things that we could do with AI, but there are some other things that we can’t do, and is very important to know is your project is possible.
Collect data
ML models tend to work very well when there are a lot of data available to train the model. So this is a very important step, collect the data to your project. As I said above, the type of data you need will be designate by the type of problem you are trying to solve.
If you are doing a project just to learn or train your ML skills, this could be a very easy step, since there are a lot of interesting datasets out there. If you are doing a very customizable project or you are doing a project for your company, this step can get a bit complex, since the data will have to be collected by you. And here, you need to think about creative ways to collect data.
For example, if you have an online store, and you are trying to solve a store problem, and in order to solve the problem you need to get a lot of data from the users, it will be easy, because the users can generate all that data for you just by using your store.
There are tons of ways of collecting data to solve specific problems. And first you need to know what are you trying to do before collecting data.
Train the model
Before start training the model, you need to pick a model, an algorithm, so you need to think hardly about your problem and the best way to solve it, because that will give you the key to know which algorithm or model to use.
Now that we have the data, it’s time to put it to work, and that’s done by training the model. In this step, we need to iterate a lot, because in the first training our model might not be good enough. So we need to make small adjustments and see how much it improves, and iterate on this until the model is good enough.
Also, we need to test the behavior of the model with data that it has never seen. And here and surge two problems, either the new data isn’t good enough or there is a problem with the model.
If the model is good enough, with the training data and with the new data, the next step is to deploy the model.
Deploy the model
If the project was for educational purposes, you may not want or need to deploy it. But if the project was for other purposes, it may need to be done.
Once deployed, there are important things to do in order to improving the model. In this step, the model will be working with a lot of data that it has never seen. So, we can collect this new data and incorporate it to the training dataset. And with new data in the dataset, we can retrain, maintain and improve the model.
And that’s the workflow of a ML project. The best way to solve a big problem (such as an ML project) is to break it down into smaller parts or steps, and that’s what the workflow of a Machine Learning project does.
¡Thank you for reading me and see you next!
