# What was the basis for parameter choices in tutorials?

**URL:** <https://discourse.scverse.org/t/what-was-the-basis-for-parameter-choices-in-tutorials/399>\
**Category:** scvi-tools\
**Tags:** tutorials, scanvi\
**Created:** [April 11, 2022, 1:11pm UTC](https://discourse.scverse.org/t/what-was-the-basis-for-parameter-choices-in-tutorials/399 "2022-04-11T13:11:38Z")\
**Posts on this page:** 2\
**Page:** 1

<div class="post-metadata">

**Author:** ![TonyK](https://avatars.discourse-cdn.com/v4/letter/t/4491bb/32.png) [@TonyK](https://discourse.scverse.org/u/TonyK)\
**Post date:** [April 11, 2022, 1:11pm UTC](https://discourse.scverse.org/t/what-was-the-basis-for-parameter-choices-in-tutorials/399/1 "2022-04-11T13:11:38Z")

</div>

Hello,

I was wondering what the basis was for the parameters chosen in the tutorials such as “Reference mapping with scvi-tools” and “Querying the Human Lung Cell Atlas”. Specifically, the number of epochs and the training arguments:

```auto
arches_params = dict(
    use_layer_norm="both",
    use_batch_norm="none",
    encode_covariates=True,
    dropout_rate=0.2,
    n_layers=2,
)
surgery_epochs = 500
train_kwargs_surgery = {
    "early_stopping": True,
    "early_stopping_monitor": "elbo_train",
    "early_stopping_patience": 10,
    "early_stopping_min_delta": 0.001,
    "plan_kwargs": {"weight_decay": 0.0},
}

```

Were these parameters chosen from some type of search or is there some rule of thumb that was applied?

Cheers,  
Tony

---

<div class="post-metadata">

**Author:** ![pseudonym2](https://yyz1.discourse-cdn.com/flex035/user_avatar/discourse.scverse.org/pseudonym2/32/89_2.png) [@pseudonym2](https://discourse.scverse.org/u/pseudonym2)\
**Post date:** [April 29, 2022, 8:48pm UTC](https://discourse.scverse.org/t/what-was-the-basis-for-parameter-choices-in-tutorials/399/2 "2022-04-29T20:48:27Z")

</div>

I also think would be good to know.