Generate a Tensorboard projector for visualzing the embeddings

Automatically generate Tensorboard event files and launch a visualizer for it. Currently suppot zarr and npy files.

Arguments

--channel_first

Channel first for the embedding vectors. Default is True. (default value: 1)

--embeddings

Path to the feature vector embeddings (default value: PARAMETER_VALUE)

--labels

Path to the labels correspond to the embeddings. Labels should be (N,) shape or a flattened array. (default value: PARAMETER_VALUE)

--embed_zarr_key

Key to get the embeding zarr file. (default value: PARAMETER_VALUE)

--label_zarr_key

Key to get the label zarr file. (default value: PARAMETER_VALUE)

--label_img

Path to the npy image files corresponds to the embeddings. (default value: PARAMETER_VALUE)

--logdir

Output directory name in the current working directory/runs. Default is runs/tensorboard (default value: runs/tensorboard)

--k

Percentage of the features to visualize (<= 1000). Default is 0.01 (default value: 0.01)

--port

Port number to launch Tensorboard server. Default is 6006 (default value: 6006)

Please follow this link for details on how to install and run this solution.