# Validation\_split in keras policy?

**URL:** <https://forum.rasa.com/t/validation-split-in-keras-policy/19902>\
**Category:** Rasa Open Source\
**Created:** [October 16, 2019, 12:39pm UTC](https://forum.rasa.com/t/validation-split-in-keras-policy/19902 "2019-10-16T12:39:56Z")\
**Posts on this page:** 2\
**Page:** 1

<div class="post-metadata">

**Author:** ![piyush29programmer](https://dub1.discourse-cdn.com/flex013/user_avatar/forum.rasa.com/piyush29programmer/32/4050_2.png) [@piyush29programmer](https://forum.rasa.com/u/piyush29programmer)\
**Post date:** [October 16, 2019, 12:39pm UTC](https://forum.rasa.com/t/validation-split-in-keras-policy/19902/1 "2019-10-16T12:39:56Z")

</div>

Hi All,

I am using the below Keras policy while training the model.

```
policies:
  - name: KerasPolicy
    epochs: 100
    max_history: 5
    priority: 1
    batch_size: 100
    validation_split: 0.2

```

As per documentation validation\_split, 0.2 means that 80% of the data is used to train the model and the rest 20% will be used to test the model.

So the loss and acc are calculated using 20% of the dataset.

 ![image](https://europe1.discourse-cdn.com/flex013/uploads/rasa/original/2X/0/0e02a4310ec51a28115dc07dbc04a79ea83fa094.png)

Will the final model will have the 20% dataset also as we are using it for testing in the train to measure how the model performs for unseen data? if not then we are losing the 20% data and it’s a big risk? we want 100% of data (intent) to train the model.

Please tell me if my understanding is correct.

---

<div class="post-metadata">

**Author:** ![amn41](https://dub1.discourse-cdn.com/flex013/user_avatar/forum.rasa.com/amn41/32/20_2.png) [@amn41](https://forum.rasa.com/u/amn41)\
**Post date:** [October 17, 2019, 7:38am UTC](https://forum.rasa.com/t/validation-split-in-keras-policy/19902/2 "2019-10-17T07:38:59Z")

</div>

hi @piyush29programmer - this isn’t really a Rasa specific question, but the idea behind using a validation set is to make sure you don’t overfit to your training data. There are many good resources on this topic online