# Selection of train set

**URL:** <https://community.drivendata.org/t/selection-of-train-set/7591>\
**Category:** Mars Spectrometry\
**Created:** [April 4, 2022, 8:59am UTC](https://community.drivendata.org/t/selection-of-train-set/7591 "2022-04-04T08:59:03Z")\
**Posts on this page:** 3\
**Page:** 1

<div class="post-metadata">

**Author:** ![onurk83](https://avatars.discourse-cdn.com/v4/letter/o/a183cd/32.png) [@onurk83](https://community.drivendata.org/u/onurk83)\
**Post date:** [April 4, 2022, 8:59am UTC](https://community.drivendata.org/t/selection-of-train-set/7591/1 "2022-04-04T08:59:03Z")

</div>

Hi,

I have a question.

1. Do we have to use all training set(766) for every label?

2. Instead of using all zeros as 0 in the training set, can we train the label with low correlation as 0 ?

For example,

Can we use train data for “basalt” like below?

train ones =\> basalt  
train zeros =\> oxalate + sulfide (min correlation with basalt)

---

<div class="post-metadata">

**Author:** ![jayqi](https://yyz2.discourse-cdn.com/flex028/user_avatar/community.drivendata.org/jayqi/32/924_2.png) [@jayqi](https://community.drivendata.org/u/jayqi)\
**Post date:** [April 4, 2022, 2:26pm UTC](https://community.drivendata.org/t/selection-of-train-set/7591/2 "2022-04-04T14:26:47Z")

</div>

Hi @onurk83,

You may train your model and prepare your training data in any way you choose when using the train and validation set data. There are no restrictions on how you use the provided train and validation sets.

Your model needs to be able to perform _inference_ on new data one sample at a time and make predictions on the target label classes in accordance with the submission format.

---

<div class="post-metadata">

**Author:** ![onurk83](https://avatars.discourse-cdn.com/v4/letter/o/a183cd/32.png) [@onurk83](https://community.drivendata.org/u/onurk83)\
**Post date:** [April 4, 2022, 2:29pm UTC](https://community.drivendata.org/t/selection-of-train-set/7591/3 "2022-04-04T14:29:46Z")

</div>

> [@jayqi](#):
>
> You may train your model and prepare your training data in any way you choose when using the train and validation set data. There are no restrictions on how you use the provided train and validation sets.
> 
> Your model needs to be able to perform _inference_ on new data one sample at a time and make predictions on the target label classes in accordance with the submission format.

I understood. Thanks for your explanation.