Tensorboard: Accessing tensor-objects in the tags of event_accumulator

Question

I'm trying to access some data stored in a Tensorboard-file. I would prefer not opening it with the Tensorboard GUI in my browser but directly open it in a python-script to be able to do further calculations.

My current code:

file = # here, filename is provided
ea = event_accumulator.EventAccumulator(event_file[0])
ea.Reload()
print(ea.Tags())

Now, my tags (ea.Tags()) are sth. like this:

{'histograms': [], 'scalars': [], 'tensors': ['observable1', 'observable2' ], ...}

First thing that's interesting is that my data of the observables are not saved in 'scalars' but in 'tensors'. How can I access these observables now? I would expect that each of the two oberservables gives an array/ a list of values (that's what I'm interested in) and probably also some tensor-related data like shapes, datatype etc.

I've already tried out to access the tensors using

x=ea.Tensors("observable1")
print(x[0])

or similar, but I got stuck in there as the output is sth. like this:

TensorEvent(wall_time=1234567890.987654, step=123, tensor_proto=dtype: DT_FLOAT
tensor_shape {
}
tensor_content: "\123Y\123@"
)

and x seems to have a fixed length of 10 which was, in somehow, unexpected to me. Does somebody have an idea? All the explanations I found on the internet are dealing only about the scalars in a Tensorboard-file

Answer 1

Provided that it has been previously written like

from torch.utils.tensorboard import SummaryWriter
tensorboard_writer=SummaryWriter(log_dir=logpath)
tensorboard_writer.add_scalar("loss_val", loss, parameter)

this example will extract the score:

from tensorboard.backend.event_processing.event_accumulator import EventAccumulator
    
event_acc = EventAccumulator(logpath, size_guidance={"scalars": 0})
event_acc.Reload()

for scalar in event_acc.Tags()["scalars"]:
    w_times, step_nums, vals = zip(*event_acc.Scalars(scalar))

Perhaps writing the scalars wasn't successful?