# %%
import pandas as pd
import numpy as np

# %%
data_path = '../../data_import/exports/data_mapping_mdm.csv'
full_df = pd.read_csv(data_path, skipinitialspace=True)
mdm_list = sorted(list((set(full_df['thing'] + full_df['property']))))


# %%
fold = 5
file_path = f'../../train/classification_bert_complete_desc_unit/classification_prediction/exports/result_group_{fold}.csv'
df_bert = pd.read_csv(file_path)
# %%
file_path = f'../../train/mapping_t5_complete_desc_unit/mapping_prediction/exports/result_group_{fold}.csv'
# file_path = f'../../train/mapping_t5-base_desc_unit/mapping_prediction/exports/result_group_{fold}.csv'
df_t5 = pd.read_csv(file_path)
df_t5 = df_t5[df_t5['MDM']].reset_index(drop=True)
df_t5['class_prediction'] = (df_t5['p_thing'] + df_t5['p_property'])
df_t5['in_vocab'] = df_t5['class_prediction'].isin(mdm_list)

# %%
df_t5['bert_prediction'] = df_bert['class_prediction']
df_bert['t5_prediction'] = df_t5['class_prediction']
# %%
bert_correct = (df_bert['thing'] + df_bert['property']) == df_bert['class_prediction']
# %%
t5_correct = (df_t5['thing'] + df_t5['property']) == (df_t5['p_thing'] + df_t5['p_property'])

# %%
sum(t5_correct)/len(t5_correct)

# %%
# replace t5 not in vocab with bert values
t5_correct_modified = t5_correct.copy()
condition = ~df_t5['in_vocab']
t5_correct_modified[condition] = np.array(bert_correct[condition])

# %%
# new replacement correctness
sum(t5_correct_modified)/len(t5_correct_modified)
# %%
# when bert is correct and t5 is wrong
cond_mask = bert_correct & (~t5_correct)
print(sum(cond_mask))
print(df_t5[cond_mask].to_string())
# %%
# when bert is wrong and t5 is correct
cond_mask = (~bert_correct) & (t5_correct)
print(sum(cond_mask))
print(df_bert[cond_mask].to_string())



# %%
# when both are wrong
cond_mask = (~bert_correct) & (~t5_correct)
print(sum(cond_mask))


# %%