Optimize predict() to remove O(n²) dataframe iteration and filtering

[midaa1]

Description

The predict() function currently builds the output dictionary using:

for index, row in df_data.iterrows():

Inside the loop, it repeatedly filters the entire dataframe to collect values for each (True X, True Y) pair:

df_data[
    (df_data["True X"] == row["True X"]) &
    (df_data["True Y"] == row["True Y"])
]

This causes:

   - O(n²) time complexity

Repeated work for duplicated (True X, True Y) combinations

Significant slowdown on medium/large datasets

Unnecessary memory allocations

Root Cause

The algorithm iterates row-by-row while performing full-dataframe filtering for each iteration, despite the fact that grouping by truth coordinates is already available.

Proposed Solution

Refactor the dictionary-building logic to:

Group data once using:

df_data.groupby("True XY'')

Iterate over each group only once.

Extract:

• Predicted X

• Predicted Y

• Precomputed precision_xy

• Precomputed accuracy_xy

Build the nested output dictionary directly from groups.

Expected Benefits

• Reduce complexity from O(n²) → O(n)

• Eliminate duplicated computations

• Improve scalability for large datasets

• Cleaner and more maintainable code

Acceptance Criteria

• No iterrows() usage in this section.

• No dataframe filtering inside loops.

• Output structure remains identical.

• Metrics (PrecisionSD, Accuracy) unchanged.

• Performance improvement confirmed on larger datasets.