Regression head fine tuning

Regression head fine tuning#

We have built an example for training a regression head on top of feature maps extracted from the frozen Clay encoder.

All the code for this example can be found in the regression finetuning folder of this repository.

Download data#

The data comes as multifile zip, it can be downloaded from the BioMassters huggingface repository. Grab a coffee, this is about 250GB in size.

The next step is to unzip training data. The data comes in a multi-file zip archive. So it needs to be unzipped using a library that can handle the format. 7z works quite well in this case. Grabb another coffee, this will take a while.

sudo apt install p7zip-full

Extract train feature#

7z e -o/home/tam/Desktop/biomasters/train_features/ /datadisk/biomasters/raw/train_features.zip

Should look something like this

7-Zip [64] 16.02 : Copyright (c) 1999-2016 Igor Pavlov : 2016-05-21
p7zip Version 16.02 (locale=en_US.UTF-8,Utf16=on,HugeFiles=on,64 bits,16 CPUs Intel(R) Core(TM) i7-10875H CPU @ 2.30GHz (A0652),ASM,AES-NI)

Scanning the drive for archives:
1 file, 10247884383 bytes (9774 MiB)

Extracting archive: /datadisk/biomasters/raw/train_features.zip
--
Path = /datadisk/biomasters/raw/train_features.zip
Type = zip
Physical Size = 10247884383
Embedded Stub Size = 4
64-bit = +
Total Physical Size = 149834321503
Multivolume = +
Volume Index = 13
Volumes = 14

Everything is Ok

Folders: 1
Files: 189078
Size:       231859243932
Compressed: 149834321503

Extract train AGBM#

7z e -o/home/tam/Desktop/biomasters/train_agbm/ /datadisk/biomasters/raw/train_agbm.zip

Should look something like this

7-Zip [64] 16.02 : Copyright (c) 1999-2016 Igor Pavlov : 2016-05-21
p7zip Version 16.02 (locale=en_US.UTF-8,Utf16=on,HugeFiles=on,64 bits,16 CPUs Intel(R) Core(TM) i7-10875H CPU @ 2.30GHz (A0652),ASM,AES-NI)

Scanning the drive for archives:
1 file, 575973495 bytes (550 MiB)

Extracting archive: /datadisk/biomasters/raw/train_agbm.zip
--
Path = /datadisk/biomasters/raw/train_agbm.zip
Type = zip
Physical Size = 575973495

Everything is Ok

Folders: 1
Files: 8689
Size:       2280706098
Compressed: 575973495

Extract test features#

7z e -o/home/tam/Desktop/biomasters/test_features/ /datadisk/biomasters/raw/test_features_splits.zip

Should look something like this

7-Zip [64] 16.02 : Copyright (c) 1999-2016 Igor Pavlov : 2016-05-21
p7zip Version 16.02 (locale=en_US.UTF-8,Utf16=on,HugeFiles=on,64 bits,16 CPUs Intel(R) Core(TM) i7-10875H CPU @ 2.30GHz (A0652),ASM,AES-NI)

Scanning the drive for archives:
1 file, 6912625480 bytes (6593 MiB)

Extracting archive: /datadisk/biomasters/raw/test_features_splits.zip
--
Path = /datadisk/biomasters/raw/test_features_splits.zip
Type = zip
Physical Size = 6912625480
Embedded Stub Size = 4
64-bit = +
Total Physical Size = 49862298440
Multivolume = +
Volume Index = 4
Volumes = 5

Everything is Ok

Folders: 1
Files: 63348
Size:       78334396224
Compressed: 49862298440

Extract test AGBM#

7z e -o/home/tam/Desktop/biomasters/test_agbm/ /datadisk/biomasters/raw/test_agbm.tar

Should look something like this

7-Zip [64] 16.02 : Copyright (c) 1999-2016 Igor Pavlov : 2016-05-21
p7zip Version 16.02 (locale=en_US.UTF-8,Utf16=on,HugeFiles=on,64 bits,16 CPUs Intel(R) Core(TM) i7-10875H CPU @ 2.30GHz (A0652),ASM,AES-NI)

Scanning the drive for archives:
1 file, 729766400 bytes (696 MiB)

Extracting archive: /datadisk/biomasters/raw/test_agbm.tar
--
Path = /datadisk/biomasters/raw/test_agbm.tar
Type = tar
Physical Size = 729766400
Headers Size = 1421312
Code Page = UTF-8

Everything is Ok

Folders: 1
Files: 2773
Size:       727862586
Compressed: 729766400

Prepare data#

This will take the average of all timesteps available for each tile. The time steps for Sentinel-2 are not complete, not all months are provided for all tiles. In addtion, the Clay model does not take time series as input. So aggregating the time element is simplifying but ok for the purpose of this example.

In addition, we skip the one orbit because it nodata most of the time

Prepare training features#

python finetune/regression/preprocess_data.py \
  --features=/home/tam/Desktop/biomasters/train_features/ \
  --cubes=/home/tam/Desktop/biomasters/train_cubes/ \
  --processes=12 \
  --sample=1 \
  --overwrite

Prepare test features#

python finetune/regression/preprocess_data.py \
  --features=/home/tam/Desktop/biomasters/test_features/ \
  --cubes=/home/tam/Desktop/biomasters/test_cubes/ \
  --processes=12 \
  --sample=1 \
  --overwrite

Train the model#

To ensure that the imports work properly, ensure that the root of the repository is in the python path before running the script.

python -m finetune.regression.regression fit --config configs/regression_biomasters.yaml
Contents