def point_test_grad_impl(): # check gradient implementation rng = nprs(42) test_ob = rng.uniform(size=(4, )) test_act = rng.uniform(size=(4, )) test_theta = rng.uniform(size=(4, 5)) # Check that the shape matches assert point_get_grad_logp_action(test_theta, test_ob, test_act).shape == test_theta.shape gradient_check( lambda x: point_get_logp_action(x.reshape(test_theta.shape), test_ob, test_act), lambda x: point_get_grad_logp_action(x.reshape(test_theta.shape), test_ob, test_act).flatten(), test_theta.flatten())
def point_test_grad_impl(): # check gradient implementation rng = nprs(42) test_ob = rng.uniform(size=(4,)) test_act = rng.uniform(size=(4,)) test_theta = rng.uniform(size=(4, 5)) # Check that the shape matches assert point_get_grad_logp_action( test_theta, test_ob, test_act).shape == test_theta.shape gradient_check( lambda x: point_get_logp_action( x.reshape(test_theta.shape), test_ob, test_act), lambda x: point_get_grad_logp_action( x.reshape(test_theta.shape), test_ob, test_act).flatten(), test_theta.flatten() )