def errorCurvePotinFiveByAnnotation(talkSeries):
ma = MimiAnalyser('data/test_data_extremelyPure.csv')
errorV1 = pd.Series()
countV1 = pd.Series()
errorA1 = pd.Series()
countA1 = pd.Series()
for talk, sentencesInTalk in talkSeries.iteritems():
dfAudioV = ma.get_V_FrameByTalk(talk, 'Audio')
dfAudioA = ma.get_A_FrameByTalk(talk, 'Audio')
aveAudioV = dfAudioV.apply(np.mean, axis=0)
aveAudioA = dfAudioA.apply(np.mean, axis=0)
dfTextV = ma.get_V_FrameByTalk(talk, 'text')
dfTextA = ma.get_A_FrameByTalk(talk, 'text')
aveTextV = dfTextV.apply(np.mean, axis=0)
aveTextA = dfTextA.apply(np.mean, axis=0)
for i in sentencesInTalk:
d = pointFive(aveAudioV[i])
eV = errorV1
cV = countV1
if eV.get(d) == None:
cV.at[d] = 1
eV.at[d] = np.abs(d - pointFive(aveTextV[i]))
else:
cV.at[d] += 1
eV.at[d] += np.abs(d - pointFive(aveTextV[i]))
for i in sentencesInTalk:
d = pointFive(aveAudioA[i])
eA = errorA1
cA = countA1
if eA.get(d) == None:
cA.at[d] = 1
eA.at[d] = np.abs(d - pointFive(aveTextA[i]))
else:
cA.at[d] += 1
eA.at[d] += np.abs(d - pointFive(aveTextA[i]))
eV = eV.div(cV).sort_index()
cV = cV.sort_index()
eA = eA.div(cA).sort_index()
cA = cA.sort_index()
errorArray = [eV, eA]
countArray = [cV, cA]
for index, countSeries in enumerate(countArray):
for label, count in countSeries.iteritems():
if count < 1:
countArray[index] = countArray[index].drop(label)
errorArray[index] = errorArray[index].drop(label)
return [errorArray[0], errorArray[1], countArray[0], countArray[1]]
import numpy as np
import pandas as pd
import matplotlib.pyplot as plt
from mimicry_analyser import MimiAnalyser
ma = MimiAnalyser('data/test_data_filtered.csv')
ma.plotTalkError(3, 'text')
ma.plotTalkError(3, 'Fumoji')
import numpy as np
import pandas as pd
import matplotlib.pyplot as plt
from mimicry_analyser import MimiAnalyser
ma = MimiAnalyser('data/test_data_filtered.csv')
THE_TALK = 13
dfAudioV = ma.get_V_FrameByTalk(THE_TALK, 'Audio')
dfAudioA = ma.get_A_FrameByTalk(THE_TALK, 'Audio')
aveAudioV = dfAudioV.apply(np.mean, axis=0)
aveAudioA = dfAudioA.apply(np.mean, axis=0)
dfTextV = ma.get_V_FrameByTalk(THE_TALK, 'text')
dfTextA = ma.get_A_FrameByTalk(THE_TALK, 'text')
aveTextV = dfTextV.apply(np.mean, axis=0)
aveTextA = dfTextA.apply(np.mean, axis=0)
dfFumojiV = ma.get_V_FrameByTalk(THE_TALK, 'Fumoji')
dfFumojiA = ma.get_A_FrameByTalk(THE_TALK, 'Fumoji')
aveFumojiV = dfFumojiV.apply(np.mean, axis=0)
aveFumojiA = dfFumojiA.apply(np.mean, axis=0)
data = {
"aveAudioV": aveAudioV.values,
"aveFumojiV": aveFumojiV.values,
"aveTextV": aveTextV.values
}
import numpy as np
import pandas as pd
import matplotlib.pyplot as plt
from mimicry_analyser import MimiAnalyser
ma = MimiAnalyser('data/test_data_pure.csv')
THE_TALK = 12
dfAudioV = ma.get_V_FrameByTalk(THE_TALK, 'audio')
dfAudioA = ma.get_A_FrameByTalk(THE_TALK, 'audio')
averageV = dfAudioV.apply(np.mean, axis=0)
averageA = dfAudioA.apply(np.mean, axis=0)
quadrantV = ma.getQuadrant(dfAudioV)
quadrantA = ma.getQuadrant(dfAudioA)
averageQV = quadrantV.apply(np.mean, axis=0)
averageQA = quadrantA.apply(np.mean, axis=0)
correctionQV = []
correctionQA = []
for index, r in dfAudioV.iterrows():
correctionQV.append(ma.quadrantCorrection(averageQV, r))
for index, r in dfAudioA.iterrows():
correctionQA.append(ma.quadrantCorrection(averageQA, r))
quadrantCorrectionV = pd.Series(correctionQV,
index=ma.user_id[THE_TALK - 2]).sort_values()
quadrantCorrectionA = pd.Series(correctionQA,
def errorCurvePotinFive():
ma = MimiAnalyser('data/test_data_extremelyPure.csv')
errorV1 = pd.Series()
countV1 = pd.Series()
errorV2 = pd.Series()
countV2 = pd.Series()
errorA1 = pd.Series()
countA1 = pd.Series()
errorA2 = pd.Series()
countA2 = pd.Series()
for talk in ma.talk_id:
dfAudioV = ma.get_V_FrameByTalk(talk, 'Audio')
dfAudioA = ma.get_A_FrameByTalk(talk, 'Audio')
aveAudioV = dfAudioV.apply(np.mean, axis=0)
aveAudioA = dfAudioA.apply(np.mean, axis=0)
dfTextV = ma.get_V_FrameByTalk(talk, 'text')
dfTextA = ma.get_A_FrameByTalk(talk, 'text')
aveTextV = dfTextV.apply(np.mean, axis=0)
aveTextA = dfTextA.apply(np.mean, axis=0)
for i in range(aveAudioV.size):
d = pointFive(aveAudioV[i])
if aveAudioA[i] < 0:
eV = errorV1
cV = countV1
elif aveAudioA[i] > 0:
eV = errorV2
cV = countV2
if eV.get(d) == None:
cV.at[d] = 1
eV.at[d] = np.abs(d - pointFive(aveTextV[i]))
else:
cV.at[d] += 1
eV.at[d] += np.abs(d - pointFive(aveTextV[i]))
for i in range(aveAudioA.size):
d = pointFive(aveAudioA[i])
if aveAudioV[i] < 0:
eA = errorA1
cA = countA1
elif aveAudioV[i] > 0:
eA = errorA2
cA = countA2
if eA.get(d) == None:
cA.at[d] = 1
eA.at[d] = np.abs(d - pointFive(aveTextA[i]))
else:
cA.at[d] += 1
eA.at[d] += np.abs(d - pointFive(aveTextA[i]))
errorV1 = errorV1.div(countV1).sort_index()
countV1 = countV1.sort_index()
errorA1 = errorA1.div(countA1).sort_index()
countA1 = countA1.sort_index()
errorV2 = errorV2.div(countV2).sort_index()
countV2 = countV2.sort_index()
errorA2 = errorA2.div(countA2).sort_index()
countA2 = countA2.sort_index()
errorArray = [errorV1, errorV2, errorA1, errorA2]
countArray = [countV1, countV2, countA1, countA2]
for index, countSeries in enumerate(countArray):
for label, count in countSeries.iteritems():
if count < 3:
countArray[index] = countArray[index].drop(label)
errorArray[index] = errorArray[index].drop(label)
data = {
'when Arousal < 0': errorArray[0],
'when Arousal > 0': errorArray[1]
}
ma.plotDict(data, 'error of Valence')
data = {
'when Valence < 0': errorArray[2],
'when Valence > 0': errorArray[3]
}
ma.plotDict(data, 'error of Arousal')
plt.legend()
plt.show()
# plt.figure(figsize=(8, 6), dpi=80)
# plt.title(title+' Arousal')
# plt.plot(group1[1], label = groupLabel1 + 'Arousal')
# plt.plot(group2[1], label = groupLabel2 + 'Arousal')
# barWidth = 0.1
# plt.bar(male[3].index - barWidth*0.5, male[3].div(100).values, barWidth, label = groupLabel1+'Count/100')
# plt.bar(female[3].index + barWidth*0.5, female[3].div(100).values, barWidth, label = groupLabel2+'Count/100')
# plt.legend()
# plt.show()
ma = MimiAnalyser('data/test_data_filtered.csv')
a = Annotation()
errorCurvePotinFive()
# series_maleSpeakers = a.getTalkSentenceByAnnotation('说话人性别','男')
# series_femaleSpeakers = a.getTalkSentenceByAnnotation('说话人性别','女')
# male = errorCurvePotinFiveByAnnotation(series_maleSpeakers)
# female = errorCurvePotinFiveByAnnotation(series_femaleSpeakers)
# genderGroupV = {'male':male[0], 'female':female[0]}
# genderGroupVCount = {'male':male[2], 'female':female[2]}
# plotGroup(genderGroupV, genderGroupVCount,'genderGroupValence')
# genderGroupA = {'male':male[1], 'female':female[1]}
# genderGroupACount = {'male':male[3], 'female':female[3]}
import numpy as np
import pandas as pd
import matplotlib.pyplot as plt
from mimicry_analyser import MimiAnalyser
ma = MimiAnalyser('data/test_data_pure.csv')
#talk : 2-13
talk_length = [52, 22, 9, 11, 30, 45, 4, 19, 10, 32, 43, 19]
data = pd.read_csv('data/test_data_pure.csv')
text_data = data[data['Type'] == 'text']
fumoji_data = data[data['Type'] == 'Fumoji']
audio_data = data[data['Type'] == 'Audio']
fumoji_valence_error = [{} for i in range(12)]
text_valence_error = [{} for j in range(12)]
# def array2Dic(array):
# dic = {}
# for i in array:
# dic[i[0]] = i[1]
# return dic
#
# def cal_error_abs(a,b):
# result = 0
# if(len(a) == len(b) and len(a) > 0):
# for (key,item) in a.items():
# result += np.abs(item - b[key])
# return float(result / len(a))