They will end up not doing its task properly.You decide to build another cat classifier, and its goal is to do classification as good as possible.If it uses a metric to tell how well it’s performing regression, optimizing that metric will not make the model do better on classification.Since choosing the right kind of error is crucial for a good performing machine learning model, how can we do that?Choosing the right metricOne of your job, when you’re creating a machine learning model is to tell your model on what to improve.To put it in other words, it’s your responsibility to choose the right error so the model can do it’s task properly.To choose the right kind of error, you need to know what your model’s task.The most common task in a model is regression or classification..It’s worth remembering the error of these tasks because they will show up a lot.To optimize a regression task, the error should be mse (or it’s variation).To optimize a classification task, the error should be log loss (or it’s variation).But as you go deeper in machine learning, you will find that it’s more than regression and classification (e.g. detecting object, or generating a new image).Each of these tasks will have its own error to optimize, and it’s important to optimize the right error so the model will learn to do it’s task properly.So before you ask people on the internet why your model isn’t performing well, a helpful question that you can ask yourself is:“Am I optimizing the right error?”This post is originally published at vinliao.com.If you find this post helpful, please give it a clap and share it with your friends..I would love to hear your thoughts, leave your comments below!. More details