Inference Command

Command examples used to inference models

Before running the commands, we must make sure the model weights are organized.

• If you followed the previous training steps, the model weights will be located in the specified output_dir of the YAML. They will already be in the correct format, so you can directly run the commands at the bottom of the page.

• If you are using other pre-trained weights, find the folder/directory in which they are stored and follow the steps below.

The folder (e.g. home/data/MODEL_WEIGHTS) should include files like:

encoder_scaler.pkl, dbn.ckpt, encoder.ckpt, deep_cluster.ckpt, heir_fc.ckpt

It also should include two .npy files, which, if missing, can be created with:

touch home/data/MODEL_WEIGHTS/train_data.indices.npy  ## replace with your path
touch home/data/MODEL_WEIGHTS/train_data.npy

Now modify the directory structure to look like this:

The script expects the files in specific folders, so it is important to not skip this step.

• Tip: Use the mkdir and mv commands to create this structure.

Now we can run our commands:

cd <path_to_SIT_FUSE>/SIT_FUSE/src/sit_fuse/inference

python3 generate_output.py -y <path_to_yaml>
# Same YAML as in the previous steps. See Code configuration

Outputs zarr files