def test_inv_select_items_none():
idx = as_variable(np.array([[0, 1, 2, 3], [0, 1, 2, 3]], dtype=np.int32))
val = as_variable(np.array([[0.5, 3.14, 0.0, -9.9], [1.0, -1.0, 1.0,
4.0]]))
out = inverse_select_items_per_row(val, idx)
assert_allclose(out.data, np.array([[], []], dtype=np.int32))
def test_inv_select_items_less_idx():
idx = as_variable(np.array([[3, 1], [1, 3]]))
val = as_variable(np.array([[0.5, 3.14, 0.0, -9.9], [1.0, -1.0, 1.0,
4.0]]))
exp = as_variable(np.array([[0.5, 0.0], [1.0, 1.0]]))
out = inverse_select_items_per_row(val, idx)
assert_allclose(out.data, exp.data)
def log_propensity_independent(self, x, action):
xp = cuda.get_array_module(action)
pred = self._predict(x)
final_action = action
if self.k > 0 and action.shape[1] < pred.shape[1]:
all_actions = F.broadcast_to(xp.arange(0, pred.shape[1],
dtype=action.data.dtype),
pred.shape)
inv_items = inverse_select_items_per_row(all_actions, action)
items = select_items_per_row(all_actions, action)
final_action = F.concat((items, inv_items), axis=1)
pred = select_items_per_row(pred, final_action)
results = F.log_softmax(pred)
if self.k > 0:
results = results[:, :self.k]
return results