# How are others improving/working on their models after a bad round? **URL:** \ **Category:** Data Science\ **Created:** [May 20, 2021, 6:08pm UTC](https://forum.numer.ai/t/how-are-others-improving-working-on-their-models-after-a-bad-round/3364 "2021-05-20T18:08:58Z")\ **Posts on this page:** 1\ **Showing post:** 3
**Author:** ![gammarat](https://yyz1.discourse-cdn.com/flex009/user_avatar/forum.numer.ai/gammarat/32/3281_2.png) [@gammarat](https://forum.numer.ai/u/gammarat)\ **Post date:** [May 21, 2021, 2:43am UTC](https://forum.numer.ai/t/how-are-others-improving-working-on-their-models-after-a-bad-round/3364/3 "2021-05-21T02:43:18Z")
It led me to start making synthetic “training\* (I use the word loosely as I’m not using NNs) data that on a broad statistical basis is the same as the real training data, but on a narrower basis is more closely aligned with the test data. The big thing that caught my eye was that the variance in the various feature groups in the training data is significantly different from that in the test and live data. I wrote a bit about it in this thread, and what I learned from that seems to have improved my modelling. > [@Analyzing Training Data](http://forum.numer.ai/t/analyzing-training-data/3143): > > I’ve been playing around with the feature groups a bit, and two things popped out having to do with the standard deviations of the features. Referring to these plots: The numbers down the side of the first plot are feature numbers, down the side of the second plot they are simply levels. Sorry about the mess. Across the bottom are file numbers from 1 to 148. 1 to 120 are training data era, 121 to 148 are validation. In the first plot, I just took the standard deviation of… --- _[View the full topic](https://forum.numer.ai/t/how-are-others-improving-working-on-their-models-after-a-bad-round/3364)._