Update semantics WS dataset

[echo0x22]

Updated variant with additional gradient (<25%(no), 25%, 50%, 75%, <100%(yes), 100%(same))

[sedthh]

I'm going to just believe you on this one.
Btw when passing random_num it can be an issue when keys are not present in the dict or list sizes are smaller than the index given. What you can do is taking the possible values and chosing randomly from them:

part = random_dict[lang][np.random.choice(random_dict[lang].keys())]
part2 = part[np.random.choice(part.keys())]  # or part if it's a list

something like that

[echo0x22]

UPD: I couldn't think of anything better, and in general it works anyway, so...