---
title: "Evaluating on out-of-distribution samples "
description: "Cutting the dataset to train a model only on samples below 100 K, we test the model's ability to predict on materials with true Tc greater than 100 and the results are not good."
url: "https://ouro.foundation/files/mmoderwell/evaluating-on-out-of-distribution-samples"
id: "5db0e8dc-e323-40f3-b60e-57789a8e89f8"
type: "file"
---

# Evaluating on out-of-distribution samples 

File by @mmoderwell in #superconductors (all) · .png · 37.5 KB

- Asset id: `5db0e8dc-e323-40f3-b60e-57789a8e89f8`
- Page: https://ouro.foundation/files/mmoderwell/evaluating-on-out-of-distribution-samples

Cutting the dataset to train a model only on samples below 100 K, we test the model's ability to predict on materials with true Tc greater than 100 and the results are not good.

## Use it

- Over Ouro MCP: `download_asset(id="5db0e8dc-e323-40f3-b60e-57789a8e89f8")` returns the file or a download link.
- Find routes that take it: `get_compatible_routes(asset_id="5db0e8dc-e323-40f3-b60e-57789a8e89f8")`, then pass it as `input_assets={"file": "5db0e8dc-e323-40f3-b60e-57789a8e89f8"}`.
- From Python: `ouro.files.retrieve("5db0e8dc-e323-40f3-b60e-57789a8e89f8")`
