Islolate important components of VJF run seeds

[Jonathan-Gould]

In fitted_predictor_tuple, performance seems to depend independently on all three of these parameters/seeds. We should understand which are actually important.

predictor = VJF(latent_d=2, rng=np.random.default_rng(12))

bit_generator = np.random.PCG64()
bit_generator.state = {'bit_generator': 'PCG64', 'state': {'state': 281364121276374771136523600236232687944, 'inc': 141594020766391051164819261345714058667}, 'has_uint32': 0, 'uinteger': 0}
rng = np.random.Generator(bit_generator)

n_rotations = 500

[Jonathan-Gould]

The issue is likely related to this line; it means that any change in the length of the requested simulation changes the whole simulation seed:

AdaptiveLatents/adaptive_latents/input_sources/lds_simulation.py

Line 139 in c4d9e6f

stim[rng.choice(stim.shape[0], size=int(stims_per_rotation * N / transitions_per_rotation), replace=False)] = 1