.root_dir attribute isn't maintained when filtering results objects

[bw4sz]

In the endless battle about the root_dir attribute, it really trips the user up that when you filter a results object, you can't immediately visualize. I don't see an obvious solution, but I think its important to highlight that this is a poor structure.

from deepforest import main
from deepforest import visualize
import matplotlib.pyplot as plt

m = main.deepforest()
m.load_model("Weecology/deepforest-bird")

image_path="/Users/benweinstein/Downloads/example_airborne_birds/5.jpg"
predictions = m.predict_tile(image_path, patch_size=600, patch_overlap=0)
# visualize the predictions

# This works, root dir is defined in predictions.root_dir from predict_tile
visualize.plot_results(predictions)
plt.show()

high_confidence = predictions[predictions.score>0.2]

# When you filter predictions, it loses the attribute.

# This is ugly and non-pythonic. 
high_confidence.root_dir = "/Users/benweinstein/Downloads/example_airborne_birds"

visualize.plot_results(high_confidence)

[bw4sz]

Related to #1210 I wonder if we made DeepForest data class a subclass of Geopandas, we could alter the filter behavior to maintain root dir.

[ethanwhite]

Related to https://github.com/weecology/DeepForest/pull/1210 I wonder if we made DeepForest data class a subclass of Geopandas, we could alter the filter behavior to maintain root dir.

Certainly seems reasonable

[bw4sz]

Pandas has a way to do this, trying it now.

import pandas as pd

class MyDataFrame(pd.DataFrame):
    _metadata = ["custom_attribute"]

    @property
    def _constructor(self):
        return MyDataFrame

df = MyDataFrame({'data': range(10)})
df.custom_attribute = 'my_value'

# Slicing the DataFrame will preserve the custom_attribute
sliced_df = df[:5]
print(sliced_df.custom_attribute)