How to get final architecture based on the output of micro_search?

[SiZuo]

Based on the output of micro_search(num_cell=2), for example: [0 0 1 0 0 4 2 0] [0 3 0 1 1 4 0 1]

According to the explanation, each block consists of 4 numbers [index_1, op_1, index_2, op_2] and the index represent outputs of the two previous cells. So, for layer 1, based on the output [0 0 1 0 0 4 2 0], how can cells of first layer take the output of second layer as a input to the cell?

Can you based on the output mentioned above, draw the structure of final network(set the num_layer=5)?

[nott0]

Here, index_1 and index_2 can be any previous index.
op_1, op2 can be [0, 1, 2, 3, 4], corresponding to separable_conv_3x3, separable_conv_5x5, average_pooling, max_pooling, identity

The index didn't represent the layer, but it represent the node.

Let say that at first there are 2 nodes already exist as input nodes, node[0] = h[0], node[1] = h[1]

The first 4 numbers is [0 0 1 0] so node[2] = h[2] = separable_conv_3x3(h[0]) + separable_conv_3x3(h[1])

and the next 4 numbers is [0 4 2 0] so node[3] = h[3] = identity(h[0]) + separable_conv_3x3(h[2]) and so on.

[Najmeh-f]

Here, index_1 and index_2 can be any previous index. op_1, op2 can be [0, 1, 2, 3, 4], corresponding to separable_conv_3x3, separable_conv_5x5, average_pooling, max_pooling, identity

The index didn't represent the layer, but it represent the node.

Let say that at first there are 2 nodes already exist as input nodes, node[0] = h[0], node[1] = h[1]

The first 4 numbers is [0 0 1 0] so node[2] = h[2] = separable_conv_3x3(h[0]) + separable_conv_3x3(h[1])

and the next 4 numbers is [0 4 2 0] so node[3] = h[3] = identity(h[0]) + separable_conv_3x3(h[2]) and so on.

Hey, do you have any idea where can the discovered architecture of the macro search be found?