When serializing the dataset, save the np.ndarrays as np.float32 or np.int32 rather than as np.float64 or np.int64, respectively. This allows us to load the data into tensors of dtype float32, which is more than efficient to run+optimize the nn models.
Changes
[x] Save preprocessed dataset as np.float32 instead of np.float64
Description
When serializing the dataset, save the
np.ndarrays
asnp.float32
ornp.int32
rather than asnp.float64
ornp.int64
, respectively. This allows us to load the data into tensors of dtypefloat32
, which is more than efficient to run+optimize the nn models.Changes