Morphological profiling using deep learning
This projects provide tools and APIs to manipulate high-throughput images for deep learning. The dataset tools are the only ones currently implemented.
To prepare microscopy datasets for deep learning we have implemented the following steps that should be run sequentially: 1) Collect illumination statistics, 2) Compress images, and 3) Create cell location indices. Prior to these three steps, we need to create a metadata file with image locations and labels.
Any of these three steps requires a configuration file written in JSON format. With this file available for a particular dataset, you can run the dataset tools as follows:
python dataset --config=data.json metadata
python dataset --config=data.json illumination
python dataset --config=data.json compression
python dataset --config=data.json locations
These commands take some time to get your dataset ready. After that, you can launch the learning commands [under construction].
Learn a convolutional network from single cell data using the following convention:
python learning --config=learn.json training