confusion about depths of network in different stages

[NdaAzr]

Figure 3 in the PDARTS paper display depths of search networks that are 5, 11, and 17 cells for stage 1, 2 and 3 of P-DARTS respectively.

How we can get these three stages? Should we do train_search three different times with changing this parameter parser.add_argument('--layers', type=int, default=5, help='total number of layers') ?
Could you please clarify this?

[chenxin061]

It is achieved by a for loop. Please refer to the code.

The --layers only sets the initial number of layers.