---
title: Additional Information
keywords: fastai
sidebar: home_sidebar
summary: "This Notebook contains information on use of _deepflash2_."
description: "This Notebook contains information on use of _deepflash2_."
nb_path: "nbs/add_information.ipynb"
---
{% raw %}
{% endraw %}
- Images must have unique name or ID
- _0001.tif --> name/ID: 0001; img_5.png --> name/ID: img5, ...
- Corresponding masks must start with name or ID + a mask suffix
- _0001 -> 0001_mask.png (mask_suffix = "mask.png")
- _0001 -> 0001.png (masksuffix = ".png")
Learning-Rate-Finder
- Learning rate: controls how quickly or slowly a neural network model learns.
We need to set the maximum learning rate max_lr
when traing with the One-Cycle-Policy. Using the Learning Rate Finder below, a good value for max_lr
is somthing in the range of:
- one tenth of the minimum before the divergence (Minimum/10)
- when the slope is the steepest (steepest point)
- In our experiments, we found that a maximum learning rate of 5e-4 (e.g., 0.0005) yielded the best results across experiments.
Train-validation-split
n_models = 1
leads to a train-validation-split of 0.75/0.25
n_models > 1
(model ensembles) to _k-fold cross validation_