# How many predictions do we need to make?

**URL:** <https://community.drivendata.org/t/how-many-predictions-do-we-need-to-make/5617>\
**Category:** MagNet: Model the Geomagnetic Field\
**Created:** [December 17, 2020, 11:03am UTC](https://community.drivendata.org/t/how-many-predictions-do-we-need-to-make/5617 "2020-12-17T11:03:24Z")\
**Posts on this page:** 5\
**Page:** 1

<div class="post-metadata">

**Author:** ![belinda\_trotta](https://avatars.discourse-cdn.com/v4/letter/b/ba9def/32.png) [@belinda\_trotta](https://community.drivendata.org/u/belinda_trotta)\
**Post date:** [December 17, 2020, 11:03am UTC](https://community.drivendata.org/t/how-many-predictions-do-we-need-to-make/5617/1 "2020-12-17T11:03:24Z")

</div>

The rules say that the submission needs to finish within 8 hours, but it’s not specified how many predictions we need to make in this time. It would be good to know this so we can estimate whether our model will run within the time limit.

---

<div class="post-metadata">

**Author:** ![isms](https://yyz2.discourse-cdn.com/flex028/user_avatar/community.drivendata.org/isms/32/9_2.png) [@isms](https://community.drivendata.org/u/isms)\
**Post date:** [December 17, 2020, 4:31pm UTC](https://community.drivendata.org/t/how-many-predictions-do-we-need-to-make/5617/2 "2020-12-17T16:31:52Z")

</div>

Hi, great question: you’ll need to make predictions for t0 and t1 at 58,678 individual times. That’s very roughly half as many as for the training data, so if you can run the training data through in the time limit you have plenty of buffer room.

---

<div class="post-metadata">

**Author:** ![amogil](https://yyz2.discourse-cdn.com/flex028/user_avatar/community.drivendata.org/amogil/32/635_2.png) [@amogil](https://community.drivendata.org/u/amogil)\
**Post date:** [December 21, 2020, 11:19am UTC](https://community.drivendata.org/t/how-many-predictions-do-we-need-to-make/5617/3 "2020-12-21T11:19:45Z")

</div>

Hi,

Thank you for the answer.

58,678 rows in test and 8 hours limit mean that we have 8 x 60 x 60 / 58678 = 0.49 second limit for one row. Which can be hard in case we use raw data (10080 rows) for calculation complicated statistics such as binned entropy.

Do you have any plans to changing the time limit?

Thank you.

---

<div class="post-metadata">

**Author:** ![MineshJ1291](https://avatars.discourse-cdn.com/v4/letter/m/c4cdca/32.png) [@MineshJ1291](https://community.drivendata.org/u/MineshJ1291)\
**Post date:** [December 21, 2020, 12:33pm UTC](https://community.drivendata.org/t/how-many-predictions-do-we-need-to-make/5617/4 "2020-12-21T12:33:54Z")

</div>

Hi @amogil,

You can try downsampling with some aggregation before other calculation. It will help to speedup.

---

<div class="post-metadata">

**Author:** ![cszc](https://avatars.discourse-cdn.com/v4/letter/c/c68b51/32.png) [@cszc](https://community.drivendata.org/u/cszc)\
**Post date:** [December 21, 2020, 4:36pm UTC](https://community.drivendata.org/t/how-many-predictions-do-we-need-to-make/5617/5 "2020-12-21T16:36:26Z")

</div>

Hi @amogil. We do not currently have plans to up the time limit. I would suggest pre-computing statistics and caching them if possible. Best of luck!