benchmarks/identify-contrails/scripts/eval.py

import sys
import sys
import os
sys.path.append(os.path.join(os.path.dirname(__file__), "../env"))
import pandas as pd
from encode import rle_decode
import numpy as np

def dice_score(y_p, y_t, smooth=1e-6):
    i = np.sum(y_p * y_t)
    u = np.sum(y_p) + np.sum(y_t)
    score = (2 * i + smooth)/(u + smooth)
    return score

def get_score(submission_folder = "../env"):
    submission_path = os.path.join(submission_folder, "submission.csv")
    submission = pd.read_csv(submission_path,  index_col='record_id')
    ids = os.listdir(os.path.join(os.path.dirname(__file__), "test_answer"))
    total_score = 0
    for id in ids:
        pred = submission.loc[int(id), 'encoded_pixels'] 
        pred = rle_decode(pred)
        score = dice_score(pred, np.load(f"{os.path.dirname(__file__)}/test_answer/{id}/human_pixel_masks.npy")[:,:,0])
        total_score += score

    return total_score/len(ids)

if __name__ == "__main__":
    print(get_score())