# %%
import pandas as pd

# following code computes final mapping + classification accuracy
# %%
def run(fold):
    data_path = f'../relevant_class/binary_classifier_desc_unit/classification_prediction/exports/result_group_{fold}.csv'
    df = pd.read_csv(data_path, skipinitialspace=True)
    p_mdm = df['p_mdm']

    # data_path = f'../train/mapping_t5_complete_desc_unit_name/mapping_prediction/exports/result_group_{fold}.csv'
    data_path = f'../train/modified_t5_decoder_4_layers/mapping_prediction/exports/result_group_{fold}.csv'
    df = pd.read_csv(data_path, skipinitialspace=True)
    actual_mdm = df['MDM']

    thing_correctness = df['thing'] == df['p_thing']
    property_correctness = df['property'] == df['p_property']
    answer = thing_correctness & property_correctness

    # if is non-MDM -> then should be unmapped
    # if is MDM -> then should be mapped correctly

    # out of correctly predicted relevant data, how many are mapped correctly?
    correct_positive_mdm_and_map = sum(p_mdm & actual_mdm & answer)

    # number of correctly predicted non-relevant data
    correct_negative_mdm = sum(~(p_mdm) & ~(actual_mdm))

    overall_correct = (correct_positive_mdm_and_map + correct_negative_mdm)/len(actual_mdm)
    print(overall_correct)
# %%
for fold in [1,2,3,4,5]:
    run(fold)