Set epoch_seed_change attribute on SimulationDataset

[srstevenson]

Set epoch_seed_change attribute on SimulationDataset

This was added to the StreamingDataset which the SimulationDataset inherits, so also needed to be added here. Without this, the code attempts to access the missing attribute when running a simulation:

AttributeError: 'SimulationDataset' object has no attribute 'epoch_seed_change'
Traceback:
File "/home/scott/projects/streaming/.venv/lib64/python3.12/site-packages/streamlit/runtime/scriptrunner/exec_code.py", line 88, in exec_func_with_error_handling
    result = func()
             ^^^^^^
File "/home/scott/projects/streaming/.venv/lib64/python3.12/site-packages/streamlit/runtime/scriptrunner/script_runner.py", line 579, in code_to_exec
    exec(code, module.__dict__)
File "/home/scott/projects/streaming/simulation/interfaces/sim_ui.py", line 409, in <module>
    submit_jobs(shuffle_quality, dataset, time_per_sample, node_internet_bandwidth,
File "/home/scott/projects/streaming/simulation/interfaces/sim_ui.py", line 110, in submit_jobs
    for output in gen_sim:
                  ^^^^^^^
File "/home/scott/projects/streaming/simulation/core/main.py", line 110, in simulate
    samples_per_node = dataset.get_samples_per_node(epoch, 0)
                       ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/home/scott/projects/streaming/simulation/core/sim_dataset.py", line 367, in get_samples_per_node
    partition = generate_work(self.batching_method, self, self.world, epoch, sample_in_epoch)
                ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/home/scott/projects/streaming/streaming/base/batching/__init__.py", line 45, in generate_work
    return get(dataset, world, epoch, sample_in_epoch)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/home/scott/projects/streaming/streaming/base/batching/random.py", line 49, in generate_work_random_batching
    shuffle_units, small_per_big = dataset.resample_streams(epoch)
                                   ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/home/scott/projects/streaming/streaming/base/dataset.py", line 878, in resample_streams
    epoch, self.epoch_seed_change)
           ^^^^^^^^^^^^^^^^^^^^^^

Closes #831

[snarayan21]

Hey @srstevenson, we actually don't want to call StreamingDataset's constructor here -- SimulationDataset is meant to run on single process unlike StreamingDataset so there's some logic that's different between the two init methods. The epoch_seed_change attribute not being present is because when that attribute was added to StreamingDataset it was not also added to SimulationDataset. Admittedly, this isn't the best design pattern. But yeah the immediate solution would be to copy over that logic from StreamingDataset to SimulationDataset.

[srstevenson]

Thanks, @snarayan21, that makes sense. I've marked this PR as draft for now, and will update it to set epoch_seed_change without calling the parent constructor.

[srstevenson]

@snarayan21 I've updated this to just set epoch_seed_change with the same logic as in the parent class, PTAL.