Allow image bytes type during preprocessing

[vijayi1]

In image_feature.py, image bytes instance type is handled by all the read_image routines, except for preprocessing.
added the same bytes instance logic to the _finalize_preprocessing function.
tested the mnist example, with image paths as well as image bytes objects.

[alexsherstinsky]

LGTM -- @vijayi1 -- thank you very much for this enhancement!

[github-actions]

Unit Test Results

  6 files  ±0    6 suites  ±0   13m 49s ⏱️ - 3m 14s
12 tests ±0    7 ✔️  -   2    5 💤 +  2  0 ❌ ±0 
60 runs  ±0  30 ✔️  - 12  30 💤 +12  0 ❌ ±0 

Results for commit c98457c. ± Comparison against base commit c09d5dc.

This pull request skips 2 tests.

tests.regression_tests.benchmark.test_model_performance ‑ test_performance[ames_housing.gbm.yaml]
tests.regression_tests.benchmark.test_model_performance ‑ test_performance[mercedes_benz_greener.gbm.yaml]

[arnavgarg1]

Thanks for the PR! Is it possible to add a simple test?

[vijayi1]

I used the following on examples/mnist/ and trained with the df -

    import pandas as pd
    
    df = mnist.load()
    
    image_bytes = []
    for index, row in df.iterrows():
        img_path = row['image_path']
        f = open(img_path, mode="rb")
        img_bytes = f.read()
        f.close()
        image_bytes.append(img_bytes)
    
    df_bytes = pd.DataFrame.from_dict({'image_path':image_bytes})
    
    # replace image file paths with image bytes
    df = df.drop(['image_path'],axis=1)
    df = pd.concat([df, df_bytes], axis=1)
    
    #print(df.head())
    #df.to_parquet("temp.parquet")

[arnavgarg1]

Sounds good!