45 points by mlnerd 1 year ago flag hide 9 comments
user1 1 year ago next
I usually start by defining the problem and gathering data. Then I do exploratory data analysis, feature engineering, and model training in separate stages. This helps me understand the data and the model's performance better.
user2 1 year ago next
That's a good approach. I also like to version control my code and data using Git and DVC. This way, I can reproduce my results and collaborate with others easily.
user1 1 year ago next
I agree, version control is essential in ML projects. I also use MLflow for tracking experiments and packaging my models for deployment.
user3 1 year ago prev next
I usually follow a Jupyter notebook for my ML projects, starting from data preprocessing, EDA, and then model building and evaluation. It helps me to keep track of the progress and also allows me to share my work with others easily.
user2 1 year ago next
I prefer using a proper IDE for ML projects. Jupyter notebooks can be great for experimentation, but they can get messy and hard to manage when the project becomes bigger.
user3 1 year ago next
That's true, but there are some tools like Jupyter Lab and Google Colab that can help with managing larger notebooks. Anyways, the most important thing is to keep the code modular, maintainable and version-controlled.
user4 1 year ago prev next
I use a modular design approach, where I create different modules for data preprocessing, feature engineering, model training, and evaluation. This way, I can easily test, debug and reuse code.
user1 1 year ago next
That's a great approach, modularity makes it easier to scale up and maintain the project. Do you use any dependency management tools like pipenv or conda?
user4 1 year ago next
Yes, I use pipenv for dependency management. It helps me to create isolated environments and avoid conflicts between packages.