Markus Hinsche

Loss goes to NaN

2

For a regression task, I am using a [mid-size CNN](https://github.com/Welthungerhilfe/cgm-ml/blob/main/src/models/CNNDepthMap/CNNDepthMap-height/q3-depthmap-plaincnn-height/src/model.py) consisting of Conv and MaxPool layers in the first layers and Dense layers in the last layers. This is how...

https://github.com/Welthungerhilfe/cgm-ml/blob/main/cgmml/common/grad_cam/african_elephant.jpg has a very big file size. Let's compress this more

The warning says: "Ubuntu-latest workflows will use Ubuntu-20.04 soon. For more details, see https://github.com/actions/virtual-environments/issues/1816" Example where this appears: https://github.com/Welthungerhilfe/cgm-ml/actions/runs/430334253

Poetry uses `pyproject.toml` to replace `setup.py`, `requirements.txt`, `setup.cfg`, `MANIFEST.in`. It is used by lots of open-source projects

Example how to integrate into code and CI: https://github.com/Welthungerhilfe/cgm-ml/pull/500 Remaining work: go through the type violations resolving issues highlighted by mypy.

Change preprocess shortcoming: interpolation when resizing

1

During training, we upscale the depthmaps from size (180, 135) to (240, 180). Depthmaps contain pixels which have a values `0.` (zero), where they are invalid. This is an important...

Duplicated log lines

6

When running train.py, I get duplicated log lines: ```bash (env_p_3)$ python src/models/CNNDepthMap/CNNDepthMap-height/2021q3-depthmap-plaincnn-rgbd/src/train.py 2021-08-02 09:36:56,432 - INFO - Creating temp folder - /Users/markus/Development/cgm/cgm-ml/src/models/CNNDepthMap/CNNDepthMap-height/2021q3-depthmap-plaincnn-rgbd/src/train_util.py: line 14 INFO - Creating temp folder ......

**Problem** - The current (15fbf3a87929eb034d3afb1cd8d1f07d8494cedb) size of the repo is `153MB` and of the `.git` is `83MB`. - This can become hard to handle with a poor internet connection. **Solution:**...

`DATASET_NAME="anon-rgbd-5kscans"` `DATASET_NAME_LOCAL="anon-rgbd-5kscans-mini` The channels RGB are reversed. The name doesn't indicate that. The EDA notebooks should show samples from the dataset.

Write tests for code in src/common/

3

We have a very low test coverage: (see https://dev.azure.com/cgmorg/ChildGrowthMonitor/_build/results?buildId=449&view=logs&j=19c563d8-3dcd-57d1-cd5e-9dd946b0a29b&t=c276ab56-9c7b-560d-e018-453666437518&l=68) and we even just test a portion of the files. Especially for code in `src/common` this would help.