User bio
404 bio not found
Member since May 3, 2019
Posts:
Replies:
Very cool, thanks for sharing, Robbie!
Excellent, I love how this is end-to-end and using Interoperability!!
Open Exchange applications:
Certifications & Credly badges:
Thomas has no Certifications & Credly badges yet.
Global Masters badges:
![DC Commenter](https://community.intersystems.com/sites/default/files/badges/dc_commenter_1.png)
![DC Problem Solver](https://community.intersystems.com/sites/default/files/badges/dc_problem_solver_1.png)
![Bronze Commenter](https://community.intersystems.com/sites/default/files/badges/295959100-c193c40f-0952-4935-bea7-4317f796fbb7_0.png)
Followers:
Thomas has no followers yet.
Following:
Hi Kurro!
Thanks for your article and trying out IntegratedML. To hopefully point you in the right direction:
1. IntegratedML is not "just neural networks", but rather an autoML pipeline (see AutoML Guide) that first tests several ML methods on a subset of the data, then performs a training run using the full data using the ML method (neural networks, logistic regression, random forests, etc) that performed best on the subset of the data. In fact, by default, for regression problems like this, we only use XGBRegressor -- so in this case the method that IntegratedML uses is not a neural network at all!
2. "TRAIN MODEL" only needs to be called once per training dataset. Looping over the examples is handled inside that call.
3. This is potentially too small a dataset to produce reliable results. IntegratedML splits the data internally into training and testing subsets, so you would probably get better output if you have at least 100 random examples.
Kind Regards,
Thomas