Semi-supervised dimensionality reduction visualization with z dimension > 2

[zeroXzero]

Quoting the paper "Once the network is trained, in order to visualize the 10D learnt representation, we use a linear transformation to map the 10D representation to a 2D space such that the cluster heads are mapped to the points that are uniformly placed on a 2D circle.". With W_c = 10I we are simply adding the softmax y output to the the style parameters. I am not sure about the implementation of this linear transformation. Can you please provide some hint to implement this.

[musyoku]

Hi. I added some code. https://github.com/musyoku/adversarial-autoencoder/tree/master/run/semi-supervised/dim_reduction_and_projection

In this case ndim_z and ndim_y are the same.

Cluster head

we are simply adding the softmax y output to the the style parameters.

W_c = 10I so I implemented it as follows:

def encode_yz_representation(self, y, z):
	return 10 * y + z

Additional loss

Step.1

Generate 2D cluster head:

rad = math.radians(360 / model.ndim_y)
radius = 5
mapped_cluster_head_2d_target = np.zeros((10, 2), dtype=np.float32)
for n in range(model.ndim_y):
	x = math.cos(rad * n) * radius
	y = math.sin(rad * n) * radius
	mapped_cluster_head_2d_target[n] = (x, y)

Step.2

Optimize linear transformation:

identity = np.identity(model.ndim_y, dtype=np.float32)
mapped_head = model.linear_transformation(identity)
loss_linear_transformation = F.mean_squared_error(mapped_cluster_head_2d_target, mapped_head)
model.cleargrads()
loss_linear_transformation.backward()
optimizer_linear_transformation.update()

Result

(Training has not completed yet)

[zeroXzero]

That was quite prompt. I also did something similar for my tensorflow implementation and I am using your code as a benchmark. Thanks a lot.

[May4m]

@zeroXzero do you have the tensorflow implementation publicly available?