Section 11: The “send_message” Function

Assume that, e.g., our attention model wants patches from a specific location from a whole-slide-image.

In this case we can use the send_message function as follows:

dataloader.start()
time.sleep(10) #wait for the dataloader to load initial BigChunks.
while True:
    x, list_patients, list_smallchunks = dataloader.get()
    '''
    `x` is now a tensor of shape [`batch_size` x 3 x 224 x 224].
    `list_patients` is a list of lenght `batch_size`.
     TODO: You can use these values to, e.g., update model parameters.
     .
     .
     .
    '''
    for patient in list_patients:
        dataloader.send_message(
                            patient = patient,
                            message = {"new attention":[todo_x, todo_y]}
                            )
    if(flag_finish_running == True):
        dataloader.pause_loading()
        break

The last message is passed to the SmallChunkCollector and the BigChunkLoader as the input arguments last_message_fromroot. Please note that this is the last message associated with the Patient (i.e. self.patient in SmallChunkCollector and BigChunkLoader ). The message is not required to be a string and can be any picklable object.