[Datasets] Update read_tfrecords example

[bveeramani]

Signed-off-by: Balaji Veeramani [email protected]

Why are these changes needed?

See #28430 (comment).

Related issue number

Checks

• I've signed off every commit(by using the -s flag, i.e., git commit -s) in this PR.
• I've run scripts/format.sh to lint the changes in this PR.
• I've included any doc changes needed for https://docs.ray.io/en/master/.
• I've made sure the tests are passing. Note that there might be a few flaky tests, see the recent failures at https://flakey-tests.ray.io/
• Testing Strategy
  • Unit tests
  • Release tests
  • This PR is not tested :(

[clarkzinzow]

Nice! Although am I reading that right, that the iris.tfrecords file is 21 KB?

[bveeramani]

Nice! Although am I reading that right, that the iris.tfrecords file is 21 KB?

Yeah, you're right. Didn't check.

❯ ls python/ray/data/examples/data -la
.rw-r--r--  88k bveeramani 19 Sep 14:12 dow_jones.csv
drwxr-xr-x    - bveeramani 19 Sep 20:01 image-folders
.rw-r--r-- 4.0k bveeramani 19 Sep 14:12 iris.csv
.rw-r--r--  15k bveeramani 19 Sep 14:12 iris.json
.rw-r--r-- 3.1k bveeramani 19 Sep 14:12 iris.parquet
.rw-r--r--  22k bveeramani 23 Sep 14:07 iris.tfrecords
.rw-r--r-- 3.9k bveeramani 19 Sep 14:12 iris.tsv
.rw-r--r-- 2.5k bveeramani 19 Sep 14:12 mnist_subset.npy
drwxr-xr-x    - bveeramani 19 Sep 14:12 mnist_subset_partitioned
drwxr-xr-x    - bveeramani 19 Sep 14:12 parquet_images_mini
.rw-r--r-- 1.2k bveeramani 19 Sep 14:12 sms_spam_collection_subset.txt

To get the dataset to under 1KB, I'd have to truncate the dataset to 6 samples. Should I do that, or is it okay as-is?

[clarkzinzow]

I think it's ok as is, but does the current iris.tfrecords file have the same number of samples as e.g. the iris.csv file? I'd be surprised if so, I'd assume that the binary format would be smaller than the plain-text format.

[bveeramani]

I think it's ok as is, but does the current iris.tfrecords file have the same number of samples as e.g. the iris.csv file? I'd be surprised if so, I'd assume that the binary format would be smaller than the plain-text format.

Yeah, iris.tfrecords is created from iris.csv. Both contain 150 samples.

[clarkzinzow]

Weird 🤔 Ok I think keeping it as-is sounds good! 22 KB isn't so bad.