Hi\u000aJust thinking of use cases:\u000aSay I have a bunch of hand written doctor notes and medical test reports. I want to convert the same in the form of a proper data and then run ML models to find high risk patients. Will I be able to do that with this?\u000a\u000aAlso, does it include the complete data lifecycle in terms of all the processes from data cleaning to feature engineering and final model selection and deployment.\u000a\u000aAnd is deployment of models in built in this or do I have to deploy the model seperately? Whats the process for the same?\u000a\u000aHow much amount of data can it handle on a higher side?

Question

Waldemar_Cogniflow · Answer

Hi!\u000a\u000aHand written doctor notes could be difficult to read with great precision because typically the quality of doctor\u0027s written text is pretty bad, but you can try some of them with our OCR model and see if it can get something readable. If you want to use that to later train a model using the extracted text from the OCR model, then it does not need to be perfect and typos could be accepted by the language model and live with that.\u000a\u000aFor text data Cogniflow does some preprocessing and cleaning that you can customize (expanding the advanced option area) at the final step when creating an experiment.\u000a\u000aModel selection and deployment are addressed by Cogniflow and you do not have to worry about that. The platform trains different candidate models and hyper\u002Dparameters and it will recommend the one that gets the best optimization metric (typically F1 but you can change if you want). You always have the possibility to inspect all the candidate model and use any of them if you do not want to use the recommended one (maybe because you want to use a faster model at inference time no matter it was not the best at the accuracy level).\u000a\u000aAutomatic deployment of all candidate models is executed after training so you can use them immediately, any of the trained models through our rest API.\u000a\u000aThe amount of data depends of the experiment type, typically for text experiment you can train tens or hundreds thousands training examples, for images/audio web based solution will allow to train up to 10k \u002D 15k images. If you require more resource you can reach us and let us know, we\u0027ll be happy to help you.\u000a\u000aHope this can be helpful.\u000a\u000aHappy training!