Add __repr__ for PolymatrixGame

[oyamad]

Similar to that of NormalFormGame.

>>> g = gt.random_game((2, 3, 4))
>>> g
<2x3x4 3-player NormalFormGame of dtype float64>
>>> polymg = gt.random_polymatrix_game((2, 3, 4))
>>> polymg
<2x3x4 3-player PolymatrixGame>

@GautamsGitHub By the way, shall we enforce homogeneous dtype across players in PolymatrixGame, as in NormalFormGame?

QuantEcon.py/quantecon/game_theory/normal_form_game.py

Lines 566 to 591 in b733469

	def __init__(self, data, dtype=None):
	# data represents an array_like of Players
	if hasattr(data, '__getitem__') and isinstance(data[0], Player):
	N = len(data)
	# Check that the shapes of the payoff arrays are consistent
	# and the dtypes coincide
	shape_0 = data[0].payoff_array.shape
	dtype_0 = data[0].payoff_array.dtype
	for i in range(1, N):
	shape = data[i].payoff_array.shape
	if not (
	len(shape) == N and
	shape == shape_0[i:] + shape_0[:i]
	):
	raise ValueError(
	'shapes of payoff arrays must be consistent'
	)
	dtype = data[i].payoff_array.dtype
	if dtype != dtype_0:
	raise ValueError(
	'dtypes of payoff arrays must coincide'
	)
	self.players = tuple(data)
	self.dtype = dtype_0

[coveralls]

coverage: 93.117% (+0.001%) from 93.116%
when pulling 7ff0896 on polymg_repr
into b733469 on main.

[GautamsGitHub]

This sounds good. It is helpful because we can combine player's payoffs into vectors without worrying about there being different dtypes. I also think raising an error is better than casting everything to some datatype.

[oyamad]

Sorry, I realized that the dtype is set float whatever the dtypes of the input...

QuantEcon.py/quantecon/game_theory/polymatrix_game.py

Lines 220 to 223 in 66a4451

	matrix_builder = np.full((rows, cols), -np.inf)
	matrix_builder[:incoming.shape[0],
	:incoming.shape[1]] = incoming[:rows, :cols]
	self.polymatrix[(p1, p2)] = matrix_builder

(matrix_builder is created as an nparray of float.)