Currently we don't check logit dimension during predict. This, most commonly, can lead to issues between predict and batch_predict. For example the below will throw a non very informative error, rather than informing the user that about the incorrect dimensions of the input:
import numpy as np
from pyctcdecode.decoder import build_ctcdecoder
decoder = build_ctcdecoder(labels=["", "a", "b", "c", " "])
m = np.eye(5)
assert(decoder.decode(m) == "abc")
m = np.array([np.eye(5)])
assert(m.shape == (1, 5, 5))
decoder.decode(m)
Currently we don't check logit dimension during predict. This, most commonly, can lead to issues between predict and batch_predict. For example the below will throw a non very informative error, rather than informing the user that about the incorrect dimensions of the input: