ML approaches are easily quantifiable. -Take 5-10 small random samples from the data and keep it aside. - Use the rest of the data to train all the debating approaches, separately. - Compare the accuracy measurement metrics ( like confusion matrix for a classification problem or silhoutte score for a segmentation problem) - Test all the trained models on the sample kept aside. - Based on the last two points, decide the best ML algorithm. - No need to defend any approach in the boardroom, These two points and the numbers generated from them will speak for your ML.

To work well with data science colleagues on machine learning projects, suggest using a mix of different methods that bring together various viewpoints. Suggest regularly combining cross validation techniques and ensemble models, creating a team atmosphere where different ideas come together to form a strong, well rounded solution.

Something that worked for me a several times: Look for a decision tree algorithm and take an inspiration from it. And organize the situation into a decision tree. The root node is understanding of the problem. See if you and your colleague are on the same page or not. And gradually make your way down to other nodes and branches, to see where the disagreement is coming from. Its important to pin point the source of disagreement as well as the agreements you share. If their is enough space of agreements, discuss the solution for disagreements with open minds. Disagreements have a great potential if they are handled correctly.

Open the lines of communication to grasp differing opinions. Clearly define the project’s aims and evaluate the strengths and weaknesses of each approach, considering elements like accuracy and resource usage. If possible, conduct experiments to test the effectiveness of various methods. Consulting with external experts can also introduce new ideas. Be adaptable, incorporating elements from various approaches as needed. Focus on delivering effective results for the project, emphasizing team objectives over individual preferences. This method promotes a cooperative atmosphere, strengthens teamwork, and utilizes the diverse skills of the team for the success of the project.

When I find myself at odds with data science colleagues on ML project approaches, I focus on finding common ground. Active listening is my first step—I make an effort to understand each colleague's perspective and the reasoning behind their approach. Establishing clear project goals and success metrics that everyone agrees upon helps us evaluate different strategies objectively. I also facilitate collaborative workshops where we map each proposed approach against our project goals. This collaborative process allows us to assess the best path forward together. By fostering open communication and shared objectives, we've been able to resolve conflicts and align our team effectively.

When I'm at odds with data science colleagues on ML project approaches 1. (Just listen, don't speak) First take time to understand their perspectives by encouraging open, respectful discussion. 2. I focus on the shared goals of the project, such as improving performance or meeting stakeholder needs, to align our efforts. 3. We can test different approaches side by side using tools like A/B testing or cross-validation, comparing results with objective metrics like accuracy, precision, or recall. Use collaboration features of tools, and that allows real-time tracking.

ML models are great because they can be validated with live experiments or at least offline evaluation. Thus, the most likely scenario where you would have a disagreement with fellow scientists is when you cannot perform these validations enough times for all the approaches suggested (if you can, don’t argue; just test them all). Often this is because it takes too much resources to develop the model by different approaches. Create prototype versions of the models, with fewer features, simpler architecture, and show in offline evaluation which one is likely to be more successful. If still not clear cut, you may choose the approach with smaller risk. The most important thing is to disagree and commit once you both decide on an approach.