How does a product like this look in with regards to the trend of very reliable, well packaged open source projects, eg scikit learn?
In my experience, the challenge with many ML problems is not taking an algorithm off the shelf, but rather encoding features and making raw data sets useful.
This product is meant for a very different user base than a library like sklearn. In fact, much of our functionality leverages projects like sklearn, skimage, etc...
Basically what we're trying to do is extend a series of pre-trained models on common problems that don't require any training data from the user.
The drawback of this is that these api endpoints are less flexible than a true library, but the plus is that the user never has to deal with encoding features, or even finding, let alone cleaning raw data sets.
2 comments
[ 5.2 ms ] story [ 15.8 ms ] threadIn my experience, the challenge with many ML problems is not taking an algorithm off the shelf, but rather encoding features and making raw data sets useful.
Basically what we're trying to do is extend a series of pre-trained models on common problems that don't require any training data from the user.
The drawback of this is that these api endpoints are less flexible than a true library, but the plus is that the user never has to deal with encoding features, or even finding, let alone cleaning raw data sets.