Minor update to sort by price in show-gpus

[Vaibhav2001]

This pertains to #2863, and updates previous PR #2892

Minor update to fix the sorting issue mentioned by @concretevitamin #2892 (comment)

sky show-gpus V100 --all-regions                                                                                             ─╯
*NOTE*: for most GCP accelerators, INSTANCE_TYPE == (attachable) means the host VM's cost is not included.

GPU   QTY  CLOUD   INSTANCE_TYPE       DEVICE_MEM  vCPUs  HOST_MEM  HOURLY_PRICE  HOURLY_SPOT_PRICE  REGION
V100  1    AWS     p3.2xlarge          16GB        8      61GB      $ 3.060       $ 1.024            us-west-2
V100  1    AWS     p3.2xlarge          16GB        8      61GB      $ 3.060       $ 1.139            us-east-2
V100  1    AWS     p3.2xlarge          16GB        8      61GB      $ 3.060       $ 1.467            us-east-1
V100  1    AWS     p3.2xlarge          16GB        8      61GB      $ 3.305       $ 1.229            eu-west-1
V100  1    AWS     p3.2xlarge          16GB        8      61GB      $ 3.366       $ 0.726            ca-central-1
V100  1    AWS     p3.2xlarge          16GB        8      61GB      $ 3.589       $ 0.687            eu-west-2
V100  1    AWS     p3.2xlarge          16GB        8      61GB      $ 3.823       $ 0.971            eu-central-1
V100  1    AWS     p3.2xlarge          16GB        8      61GB      $ 4.194       $ 1.043            ap-northeast-1
V100  1    AWS     p3.2xlarge          16GB        8      61GB      $ 4.234       $ 0.469            ap-northeast-2
V100  1    AWS     p3.2xlarge          16GB        8      61GB      $ 4.234       $ 1.209            ap-southeast-2
V100  1    AWS     p3.2xlarge          16GB        8      61GB      $ 4.234       $ 1.255            ap-southeast-1
V100  4    AWS     p3.8xlarge          16GB        32     244GB     $ 12.240      $ 3.544            us-west-2
V100  4    AWS     p3.8xlarge          16GB        32     244GB     $ 12.240      $ 4.008            us-east-2
V100  4    AWS     p3.8xlarge          16GB        32     244GB     $ 12.240      $ 4.221            us-east-1
V100  4    AWS     p3.8xlarge          16GB        32     244GB     $ 13.220      $ 4.781            eu-west-1

[concretevitamin]

Thanks @Vaibhav2001. GCP requires some care; the following doesn't sort properly:

sky show-gpus --all-regions A100 --cloud gcp
sky show-gpus T4 --all-regions --cloud gcp

[concretevitamin]

Examples mentioned so far seem correct, but adding @Michaelvll to take a look at the dataframe wrangling. I'd have thought the groupby() is easier to understand than drop_duplicates(); also, seems like by removing groupby() we now have some interleaving

» sky show-gpus A100:1 --all-regions
...
A100  1    Lambda  gpu_1x_a100_sxm4  40GB        30     200GB     $ 1.100       -                  asia-northeast-1
A100  1    Lambda  gpu_1x_a100       40GB        30     200GB     $ 1.100       -                  asia-northeast-1
A100  1    Lambda  gpu_1x_a100_sxm4  40GB        30     200GB     $ 1.100       -                  asia-northeast-2
A100  1    Lambda  gpu_1x_a100       40GB        30     200GB     $ 1.100       -                  asia-northeast-2
A100  1    Lambda  gpu_1x_a100_sxm4  40GB        30     200GB     $ 1.100       -                  asia-south-1
A100  1    Lambda  gpu_1x_a100       40GB        30     200GB     $ 1.100       -                  asia-south-1
...

whereas on master

...
A100  1    Lambda  gpu_1x_a100       40GB        30     200GB     $ 1.100       -                  asia-northeast-1
A100  1    Lambda  gpu_1x_a100       40GB        30     200GB     $ 1.100       -                  asia-northeast-2
A100  1    Lambda  gpu_1x_a100       40GB        30     200GB     $ 1.100       -                  asia-south-1
A100  1    Lambda  gpu_1x_a100       40GB        30     200GB     $ 1.100       -                  australia-southeast-1
A100  1    Lambda  gpu_1x_a100       40GB        30     200GB     $ 1.100       -                  europe-central-1
A100  1    Lambda  gpu_1x_a100       40GB        30     200GB     $ 1.100       -                  europe-south-1
...

[Vaibhav2001]

Examples mentioned so far seem correct, but adding @Michaelvll to take a look at the dataframe wrangling. I'd have thought the groupby() is easier to understand than drop_duplicates(); also, seems like by removing groupby() we now have some interleaving

» sky show-gpus A100:1 --all-regions
...
A100  1    Lambda  gpu_1x_a100_sxm4  40GB        30     200GB     $ 1.100       -                  asia-northeast-1
A100  1    Lambda  gpu_1x_a100       40GB        30     200GB     $ 1.100       -                  asia-northeast-1
A100  1    Lambda  gpu_1x_a100_sxm4  40GB        30     200GB     $ 1.100       -                  asia-northeast-2
A100  1    Lambda  gpu_1x_a100       40GB        30     200GB     $ 1.100       -                  asia-northeast-2
A100  1    Lambda  gpu_1x_a100_sxm4  40GB        30     200GB     $ 1.100       -                  asia-south-1
A100  1    Lambda  gpu_1x_a100       40GB        30     200GB     $ 1.100       -                  asia-south-1
...

whereas on master

...
A100  1    Lambda  gpu_1x_a100       40GB        30     200GB     $ 1.100       -                  asia-northeast-1
A100  1    Lambda  gpu_1x_a100       40GB        30     200GB     $ 1.100       -                  asia-northeast-2
A100  1    Lambda  gpu_1x_a100       40GB        30     200GB     $ 1.100       -                  asia-south-1
A100  1    Lambda  gpu_1x_a100       40GB        30     200GB     $ 1.100       -                  australia-southeast-1
A100  1    Lambda  gpu_1x_a100       40GB        30     200GB     $ 1.100       -                  europe-central-1
A100  1    Lambda  gpu_1x_a100       40GB        30     200GB     $ 1.100       -                  europe-south-1
...

Fixed now! I am adding the sorting by intsance_type as well, because the sorting results in a reshuffling. Also I simplified the logic so we can sort regardless of --all-regions value.

[Michaelvll]

Thanks for submitting the PR @Vaibhav2001! The sort_values and drop_duplicates implementation looks good to me. Just a minor nit.

[Vaibhav2001]

@Michaelvll Hi made those changes, should be good to merge.