main.deepforest.predict_file should be able to take in a dataframe, not just a csv file.

[bw4sz]

Looking at the code base.

https://deepforest.readthedocs.io/en/latest/_modules/deepforest/main.html#deepforest.predict_file

it actually looks like dataset.TreeDataset() is the problem here, because utilities.read_file is definitely flexible enough. The docstring of predict_file should be updated and the argument name should be changed (but a deprecation warning added until 2.0), but its the dataset class, which the user really never sees that often that can be updated. I think this would still be considered a patch.

[Abhishek-kumar0503]

After read your statement it seems that utilities.read_file is reads the input CSV file and is flexible and the dataset.TreeDataset is takes the data from the CSV file and organizes it for prediction and here the issue: TreeDataset isn't flexible enough to handle different formats or structures of input data.
My solution is to Update TreeDataset to be more flexible, so i make the class of name is TreeDataset:

class TreeDataset:
      def __init__(self, csv_file, root_dir, transforms=None, train=False, column_mapping=None):
          self.root_dir = root_dir
          self.transforms = transforms
          self.train = train
          self.column_mapping = column_mapping or {
              "image_path": "image_path",
              "xmin": "xmin",
              "ymin": "ymin",
              "xmax": "xmax",
              "ymax": "ymax"
          }
       
          # Load the CSV file with flexible handling
          self.annotations = self._load_annotations(csv_file)

      def _load_annotations(self, csv_file):
          # Read the CSV file using the existing utility
          df = utilities.read_file(csv_file)
  
          # Rename columns based on column_mapping, if necessary
          df = df.rename(columns=self.column_mapping)
  
          return df 

and update the predict_file
ds = dataset.TreeDataset(csv_file=csv_file, root_dir=root_dir, transforms=None, train=False)

Is this correct?

[henrykironde]

@Abhishek-kumar0503, It looks like you're on the right track. Could you submit a PR?.

[Abhishek-kumar0503]

ya sure @henrykironde . i set up this project in my laptop. I have one query that if i change or update some function in main.py file where it will display that changes.

[vickysharma-prog]

Hi @bw4sz @ethanwhite,
Following up on this since PR #838's closing comment suggested pinging here.
I dug into the current codebase (commit d3afd3e) and found something interesting - read_file in utilities already accepts DataFrames:

def read_file(input: str | pd.DataFrame, root_dir: str | None = None, ...) -> gpd.GeoDataFrame

But FromCSVFile in datasets/prediction.py has its type hint as just str:

def __init__(self, csv_file: str, root_dir: str):

So the underlying functionality should already work with DataFrames - it's just not exposed in the type hints and docstrings.
Looked at the previous PRs too - #852 got close but stalled when tests were requested.
If this is still needed, I can submit a PR that:
Updates the type hints in FromCSVFile and predict_file
Fixes the docstrings
Adds tests for DataFrame input
Let me know if I should go ahead or if the approach has changed!

[proofbyhuman]

"Hi! I'd like to take this on as my first contribution. Following @bw4sz's guidance: update the predict_file docstring, rename the argument (keeping csv_file as a deprecated alias), and update the FromCSVFile dataset class so it accepts a DataFrame directly, since read_file already handles it downstream. One question — with the repo now on 2.1.0, should the deprecated csv_file alias target removal in 3.0? Also open to your preference on the new argument name."

[proofbyhuman]

Hi @bw4sz and maintainers! 👋

I've opened PR #1409 to address this issue. I just enabled GitHub Actions on my fork, so the CI checks should be running now.

I'm looking forward to your feedback and I'm happy to make any adjustments if needed. Thank you!