# Long 100% anomaly score spike

**URL:** <https://discourse.numenta.org/t/long-100-anomaly-score-spike/2020>\
**Category:** NuPIC\
**Created:** [March 17, 2017, 7:34am UTC](https://discourse.numenta.org/t/long-100-anomaly-score-spike/2020 "2017-03-17T07:34:22Z")\
**Posts on this page:** 2\
**Page:** 1

<div class="post-metadata">

**Author:** ![lekum](https://yyz2.discourse-cdn.com/flex030/user_avatar/discourse.numenta.org/lekum/32/1138_2.png) [@lekum](https://discourse.numenta.org/u/lekum)\
**Post date:** [March 17, 2017, 7:34am UTC](https://discourse.numenta.org/t/long-100-anomaly-score-spike/2020/1 "2017-03-17T07:34:22Z")

</div>

Hi there,

I’m creating a small demo for anomaly detection based upon the OPF cpu example, but in this case with a memory metric. I have the Jupyter notebook [here](https://github.com/lekum/htm-demo/blob/master/Infrastructure_metrics.ipynb).

I’m doing also anomaly detection and I consistently have the effect that after the 50 first samples or so I get 100% of anomaly score for another 50 samples, and the the situation stabilizes back:

 ![](https://canada1.discourse-cdn.com/flex030/uploads/numenta/original/2X/e/ebe3a4f5776d447df39a8d06dedbd42d9b211721.png)

Does anybody knows if it is somehow related to a windowing problem? Maybe I have put any wrong parameter in the model?

Thank you very much in advance

---

<div class="post-metadata">

**Author:** ![rhyolight](https://yyz2.discourse-cdn.com/flex030/user_avatar/discourse.numenta.org/rhyolight/32/3922_2.png) [@rhyolight](https://discourse.numenta.org/u/rhyolight)\
**Post date:** [March 18, 2017, 3:45pm UTC](https://discourse.numenta.org/t/long-100-anomaly-score-spike/2020/2 "2017-03-18T15:45:36Z")

</div>

Hi @lekum. I don’t think you should trust anything coming out of the HTM in this case until it has seen at least 1,000 rows of data. It needs to learn about the input for awhile to build up a model before it starts producing valuable results. I’m not surprised that it is gives high anomalies after only 50 steps.