#!/usr/bin/env python
# usage: ./evaluation_ate_absita.py RESULT_FILE GOLD_FILE


import subprocess
subprocess.run(["pip3", "install","--user", 'ndjson'])
subprocess.run(["pip3", "install","--user", 'pandas'])
subprocess.run(["pip3", "install","--user", 'numpy'])


import ndjson
import pandas as pd
import numpy as np
import sys

dataframe_predictions = pd.DataFrame()
dataframe_gold = pd.DataFrame()


with open(sys.argv[1]) as f:
    reader = ndjson.reader(f)
    for post in reader:
        df = pd.DataFrame([post], columns=post.keys())
        dataframe_predictions = pd.concat([dataframe_predictions, df], axis=0,ignore_index=True)
try:
    p_aspects = dataframe_predictions['aspects']
except:
    p_aspects = []
try:
    p_aspects_position = dataframe_predictions['aspects_position']
except:
    p_aspects_position = []
try:
    p_aspects_polarity = dataframe_predictions['polarities']
except:
    p_aspects_polarity = []
try:
    p_scores = dataframe_predictions['score']
except:
    p_scores = []

try:
    p_ids = dataframe_predictions['id_sentence']
except:
    p_ids = []
    sys.exit("Sentence IDs are missing!")

with open(sys.argv[2]) as f:
    reader = ndjson.reader(f)
    for post in reader:
        df = pd.DataFrame([post], columns=post.keys())
        dataframe_gold = pd.concat([dataframe_gold, df], axis=0,ignore_index=True)

g_aspects = dataframe_gold['aspects']
g_aspects_position = dataframe_gold['aspects_position']
g_aspects_polarity = dataframe_gold['polarities']
g_scores = dataframe_gold['score']
g_ids = dataframe_gold['id_sentence']


if (p_ids[0] != g_ids[0]):
    sys.exit("Sentences in the two files must be in the same order!")

#TASK1
precisions=[]
recalls=[]
f1s=[]


for q in range(0,len(p_aspects)):
    #Inizializza predictions
    #found_nouns = p_aspects[q]

    positions_start = []
    positions_end = []

    aspects_positions = p_aspects_position[q]
    for item in aspects_positions:
        positions_start.append(item[0])
        positions_end.append(item[1])

    #Inizializza gold
    g_starts = []
    g_ends = []

    #Inizializza gold
    gg_aspects_positions = g_aspects_position[q]
    for item in gg_aspects_positions:
        g_starts.append(item[0])
        g_ends.append(item[1])


    #Inizializza evaluation script
    ok = 0
    partial = 0
    ga = len(g_aspects[q])
    sa = len(p_aspects[q])

    f1 = 0
    pa = 0
    ra = 0

    matched = []

    #Ciclo sulle predizioni
    for k in range(0, len(positions_start)):
        #Se lo start è uguale
        if positions_start[k] in g_starts:
            ind = g_starts.index(positions_start[k])
            #se è uguale anche la fine ok, parziale altrimenti
            if g_ends[ind] == positions_end[k]:
                if ind not in matched:
                    ok = ok +1
                    matched.append(ind)
            else:
                if ind not in matched:
                    partial = partial +1
                    matched.append(ind)

        #Se end è uguale la fine
        elif positions_end[k] in g_ends:
            ind = g_ends.index(positions_end[k])
            if ind not in matched:
                partial = partial +1
                matched.append(ind)

        #Se è più corto o più lungo del gold
        else:
            for z in range(0,len(g_starts)):
                #Predizione più lunga
                if g_starts[z] < positions_start[k] and  positions_end[k] < g_ends[z]:

                    if z not in matched:
                        partial = partial +1
                        matched.append(z)
                        break
                    #Predizione più corta
                elif positions_start[k] < g_starts[z]  and g_ends[z] < positions_end[k]:
                    if z not in matched:
                        partial = partial +1
                        matched.append(z)
                        break

    #Check specific cases

    if ga== 0 and sa== 0:
        pa =1
        ra = 1
        f1 = 1
    elif sa == 0:
        pa = 0
        ra = (ok + (0.5* partial))/ga
        f1 = 0
    elif ga == 0:
        pa = (ok + (0.5* partial))/sa
        ra = 0
        f1 = 0
    else:
        pa = (ok + (0.5* partial))/sa
        ra = (ok + (0.5* partial))/ga
        if pa == 0 and ra == 0:
            f1 = 0
        else:
            f1 = (2*pa*ra)/(pa+ra)
    precisions.append(pa)
    recalls.append(ra)
    f1s.append(f1)


print("Task1 ATE:")

try:
    avg_p = np.average(precisions)
    avg_r = np.average(recalls)
    avg_f1 = np.average(f1s)
except:
    avg_p = -1
    avg_r = -1
    avg_f1 = -1

print("\tMacro-Precision: {:.5f}".format(avg_p))
print("\tMacro-Recall: {:.5f}".format(avg_r))
print("\tMacro-F1-score: {:.5f}".format(avg_f1))
print("---------------------------------------")



#TASK2
precisions=[]
recalls=[]
f1s=[]

for q in range(0,len(p_aspects_polarity)):
    #Inizializza predictions
    found_nouns = p_aspects[q]

    positions_start = []
    positions_end = []

    polarities_pred_pos = []
    polarities_pred_neg = []

    aspects_positions = p_aspects_position[q]
    for item in aspects_positions:
        positions_start.append(item[0])
        positions_end.append(item[1])

    aspects_polarity = p_aspects_polarity[q]
    for item in aspects_polarity:
        polarities_pred_pos.append(item[0])
        polarities_pred_neg.append(item[1])

    #Inizializza gold
    g_starts = []
    g_ends = []

    g_pos =[]
    g_neg = []

    #Inizializza gold
    gg_aspects_positions = g_aspects_position[q]
    for item in gg_aspects_positions:
        g_starts.append(item[0])
        g_ends.append(item[1])

    gg_aspects_polarities=g_aspects_polarity[q]
    #Inizializza gold polarities
    for item in gg_aspects_polarities:
        g_pos.append(item[0])
        g_neg.append(item[1])


    #Inizializza evaluation script
    ok = 0
    partial = 0
    ga = len(g_aspects[q])
    sa = len(found_nouns)

    f1 = 0
    pa = 0
    ra = 0

    matched = []

    #Ciclo sulle predizioni
    for k in range(0, len(positions_start)):
        #Se lo start è uguale
        if positions_start[k] in g_starts:
            ind = g_starts.index(positions_start[k])
            #se è uguale anche la fine ok, parziale altriemnti
            if g_ends[ind] == positions_end[k]:
                if ind not in matched:
                    if g_pos[ind] == polarities_pred_pos[k] and g_neg[ind] == polarities_pred_neg[k]:
                        ok = ok +1
                        matched.append(ind)
            else:
                if ind not in matched:
                    if g_pos[ind] == polarities_pred_pos[k] and g_neg[ind] == polarities_pred_neg[k]:
                        partial = partial +1
                        matched.append(ind)

        #Se end è uguale la fine
        elif positions_end[k] in g_ends:
            ind = g_ends.index(positions_end[k])
            if ind not in matched:
                if g_pos[ind] == polarities_pred_pos[k] and g_neg[ind] == polarities_pred_neg[k]:
                    partial = partial +1
                    matched.append(ind)

        #Se è più corto o più lungo del gold
        else:
            for z in range(0,len(g_starts)):
                #Predizione più lunga
                if g_starts[z] < positions_start[k] and  positions_end[k] < g_ends[z]:
                     
                    if z not in matched:
                        if g_pos[z] == polarities_pred_pos[k] and g_neg[ind] == polarities_pred_neg[k]:
                            partial = partial +1
                            matched.append(z)
                            break
                    #Predizione più corta
                elif positions_start[k] < g_starts[z]  and g_ends[z] < positions_end[k]:
                    if z not in matched:
                        if g_pos[z] == polarities_pred_pos[k] and g_neg[ind] == polarities_pred_neg[k]:
                            partial = partial +1
                            matched.append(z)
                            break

    #Check specific cases
    if ga== 0 and sa== 0:
        pa =1
        ra = 1
        f1 = 1
    elif sa == 0:
        pa = 0
        ra = (ok + (0.5* partial))/ga
        f1 = 0
    elif ga == 0:
        pa = (ok + (0.5* partial))/sa
        ra = 0
        f1 = 0
    else:
        pa = (ok + (0.5* partial))/sa
        ra = (ok + (0.5* partial))/ga
        if pa == 0 and ra == 0:
            f1 = 0
        else:
            f1 = (2*pa*ra)/(pa+ra)

    precisions.append(pa)
    recalls.append(ra)
    f1s.append(f1)


print("Task2 ABSA:")

try:
    avg_p = np.average(precisions)
    avg_r = np.average(recalls)
    avg_f1 = np.average(f1s)
except:
    avg_p = -1
    avg_r = -1
    avg_f1 = -1

print("\tMacro-Precision: {:.5f}".format(avg_p))
print("\tMacro-Recall: {:.5f}".format(avg_r))
print("\tMacro-F1-score: {:.5f}".format(avg_f1))
print("---------------------------------------")

#TASK 3
#RMSE function
def rmse(predictions, targets):
    return np.sqrt(((predictions - targets) ** 2).mean())

if len(p_scores) != 0:
    rmse_val = rmse(g_scores, p_scores)
else:
    rmse_val = -1

print("Task3 SA:")
print("\tRMSE: {:.5f}".format(rmse_val))
print("---------------------------------------")