Let’s say you want something to achieve without making your hands dirty in a code. So, it would be best if you had no-code. Not that easy to make such a thing. Microsoft tried with their own so-called CNTK, was one of the first Deep Learning tools available, epochs ago. They abandoned it since users didn’t widely accept it. They received Caffe, Theano, and TensorFlow better.
Uber tried the same thing, and they called their creation by the famous composer: Ludwig. I don’t have much experience with either of them. But, Ludwig shines with its simplicity and elegance yet providing a full-flagged toolbox for various additional settings for those interested.
Let’s say: you have a CSV file with a bunch of columns. And you don’t want to spend the time of coding reading it, splitting it to train, validation, test sets, training, etc. All you want to do is to make something quick. In such a case, all you need is to generate one YAML file, stating which are input features and which column contains outputs that you would like to make them predicted. That’s the minimum, but, yes, that’s all.
Now, you would like to make fine-tunings: feature engineering, specifying the neural network architecture you would like to use. There is no problem: a massive box with available tools is in front of you: all you need is to read the descriptions and find what you are looking for. Your YAML file will grow more and more. Still, it’s up to you. And it is easier than coding. We are not digging into discussions on how production-ready such a model would be.
Hyperparameter optimization is a big problem by itself. Ludwig solves it with a few additional keywords in the YAML file.
It is also highly appreciatable that it stays in line with the latest achievements in the area, going much more comprehensive than traditional scikit-learn. Starting with v 0.3, it supports HuggingFace’s transformers and Weights and Biases. Ludwig, respect.
Couple of useful resources, even Google would find them easily for you:
Enjoy in your no-code lunch!