Different .show behaviors in notebooks for version 0.1.5 and 0.1.6

[shyamsn97]

Describe the bug
Calling .show on a daft dataframe has different behaviors in 0.1.5 vs 0.1.6

To Reproduce
In a notebook, run this code w/ getdaft==0.1.5 and 0.1.6:

import daft
from daft import DataType
from PIL import Image
import requests
import numpy as np

images = [
    "https://raw.githubusercontent.com/Purukitto/pokemon-data.json/master/images/pokedex/hires/001.png",
]

df = daft.from_pydict({"images":images, "initial_array":np.zeros((1,512))})

@daft.udf(return_dtype=DataType.python())
def download_images(images):
    return [Image.open(requests.get(im, stream=True).raw).convert("RGB") for im in images.to_pylist()]

@daft.udf(return_dtype=DataType.python())
def arrays(column):
    return np.zeros((1, 512))

df = df.with_column("downloaded_images", download_images(daft.col('images'))).collect()
df = df.with_column("embeddings", arrays(daft.col("images"))).collect()

df.show(1)

In the code above I'm initializing a dataframe with an image url and a dummy array. Then I'm creating a column with a PIL Image that is downloaded and adding another column with a dummy array for the "embeddings"

Version 0.1.5:

Version 0.1.6:

Not sure if this is on my end w/ respect to jupyter versions / other python pkg versions. Let me know if you need any additional context, thanks!

Additional context
jupyter-lab==4.0.2
python==3.9

[jaychia]

Indeed! You actually should be able to get the displays working by using our Daft native image types instead.

I modified your code to take advantage of the new image types - using .url.download().image.decode(). Notice that PIL is no longer involved here and everything happens in Rust-land with expressions!

import daft
from daft import DataType
from PIL import Image
import requests
import numpy as np

images = [
    "https://raw.githubusercontent.com/Purukitto/pokemon-data.json/master/images/pokedex/hires/001.png",
]

df = daft.from_pydict({"images":images, "initial_array":np.zeros((1,512))})

@daft.udf(return_dtype=DataType.python())
def arrays(column):
    return np.zeros((1, 512))

df = df.with_column("downloaded_images", daft.col('images').url.download().image.decode()).collect()
df = df.with_column("embeddings", arrays(daft.col("images"))).collect()

df.show(1)

Regardless though, we should re-enable this display feature for PIL Python types! I'll work on re-enabling this feature so that if you have a python() type column with PIL images it will detect the object and display it correctly.

[shyamsn97]

awesome! Thanks for the explanation :)

[jaychia]

This should be fixed by #1078!