Hard to get into but worth it

What are the best aspects of this product?

One thing I appreciate about this tool is how constructing a machine learning model can be accomplished by getting rid of capabilities for low-level implementation. I can make use of simple Python scripts to integrate several model subblocks and models that already come prebuilt with the framework. I can take a TPU, CPU, or GPU and easily abstract it away, thanks to how this tool effectively handles implementation. I also appreciate how I don't have to think of convoluted algorithms when I'm using this tool. Handling gigabytes of data is also easier to do now thanks to how well built this tool's data ingestion pipeline is. It allows me to forget about file formatting and file access. It could be more intuitive overall, but if you want to work with applied machine learning this is an essential tool.

What aspects are problematic or could work better?

This tool's model isn't the most intuitive one out there, especially when you compare it with the other major machine learning frameworks available right now. It can be daunting for newbies and prevent them from getting into it. There is also some confusion when it comes to the variable structure of tensors since you're not clear whether you need to use plain Python types or tensors. Using it over some time makes it clearer for you, but that shouldn't be the case. This tool could be more intuitive from the jump.

What specific problems in your company were solved by this product?

Thanks to this tool, we can come up with a model that is adaptable and changeable. This is a marked improvement from how we previously built models, which would require very domain-specific knowledge. We've been able to correctly classify samples thanks to the models built from our datasets.

Are you a current user of this product?

Yes