146 lines
11 KiB
Python
146 lines
11 KiB
Python
import pandas as pd
|
|
import numpy as np
|
|
|
|
def rapids_features(sensor_data_files, time_segment, provider, filter_data_by_segment, *args, **kwargs):
|
|
|
|
conversation_data = pd.read_csv(sensor_data_files["sensor_data"])
|
|
|
|
requested_features = provider["FEATURES"]
|
|
recordingMinutes = provider["RECORDING_MINUTES"]
|
|
pausedMinutes = provider["PAUSED_MINUTES"]
|
|
# name of the features this function can compute
|
|
base_features_names = ["minutessilence", "minutesnoise", "minutesvoice", "minutesunknown","sumconversationduration","avgconversationduration",
|
|
"sdconversationduration","minconversationduration","maxconversationduration","timefirstconversation","timelastconversation","noisesumenergy",
|
|
"noiseavgenergy","noisesdenergy","noiseminenergy","noisemaxenergy","voicesumenergy",
|
|
"voiceavgenergy","voicesdenergy","voiceminenergy","voicemaxenergy","silencesensedfraction","noisesensedfraction",
|
|
"voicesensedfraction","unknownsensedfraction","silenceexpectedfraction","noiseexpectedfraction","voiceexpectedfraction",
|
|
"unknownexpectedfraction","countconversation"]
|
|
|
|
# the subset of requested features this function can compute
|
|
features_to_compute = list(set(requested_features) & set(base_features_names))
|
|
|
|
conversation_features = pd.DataFrame(columns=["local_segment"] + features_to_compute)
|
|
if not conversation_data.empty:
|
|
conversation_data = filter_data_by_segment(conversation_data, time_segment)
|
|
|
|
if not conversation_data.empty:
|
|
conversation_features = pd.DataFrame()
|
|
|
|
conversation_data = conversation_data.drop_duplicates(subset=["local_date", "local_time"], keep="first")
|
|
conversation_data[['start_ts','end_ts']] = conversation_data['timestamps_segment'].str.split(',',expand=True)
|
|
expectedMinutesDf = conversation_data[['local_segment','start_ts','end_ts']].drop_duplicates(subset=['local_segment']).set_index(['local_segment'])
|
|
expectedMinutes = (expectedMinutesDf['end_ts'].astype(int) - expectedMinutesDf['start_ts'].astype(int)) / ((60000) *(recordingMinutes + pausedMinutes))
|
|
|
|
if "minutessilence" in features_to_compute:
|
|
conversation_features["minutessilence"] = conversation_data[conversation_data['inference']==0].groupby(["local_segment"])['inference'].count()/60
|
|
|
|
if "minutesnoise" in features_to_compute:
|
|
conversation_features["minutesnoise"] = conversation_data[conversation_data['inference']==1].groupby(["local_segment"])['inference'].count()/60
|
|
|
|
if "minutesvoice" in features_to_compute:
|
|
conversation_features["minutesvoice"] = conversation_data[conversation_data['inference']==2].groupby(["local_segment"])['inference'].count()/60
|
|
|
|
if "minutesunknown" in features_to_compute:
|
|
conversation_features["minutesunknown"] = conversation_data[conversation_data['inference']==3].groupby(["local_segment"])['inference'].count()/60
|
|
|
|
if "countconversation" in features_to_compute:
|
|
conversation_features["countconversation"] = conversation_data[conversation_data["double_convo_start"] > 0].groupby(["local_segment"])['double_convo_start'].nunique()
|
|
|
|
conv_duration = (conversation_data['double_convo_end']/1000 - conversation_data['double_convo_start']/1000)/60
|
|
conversation_data = conversation_data.assign(conv_duration = conv_duration.values)
|
|
|
|
conv_totalDuration = conversation_data[(conversation_data['inference'] >= 0) & (conversation_data['inference'] < 4)].groupby(["local_segment"])['inference'].count()/60
|
|
|
|
if "silencesensedfraction" in features_to_compute:
|
|
conversation_features["silencesensedfraction"] = (conversation_data[conversation_data['inference']==0].groupby(["local_segment"])['inference'].count()/60)/ conv_totalDuration
|
|
|
|
if "noisesensedfraction" in features_to_compute:
|
|
conversation_features["noisesensedfraction"] = (conversation_data[conversation_data['inference']==1].groupby(["local_segment"])['inference'].count()/60)/ conv_totalDuration
|
|
|
|
if "voicesensedfraction" in features_to_compute:
|
|
conversation_features["voicesensedfraction"] = (conversation_data[conversation_data['inference']==2].groupby(["local_segment"])['inference'].count()/60)/ conv_totalDuration
|
|
|
|
if "unknownsensedfraction" in features_to_compute:
|
|
conversation_features["unknownsensedfraction"] = (conversation_data[conversation_data['inference']==3].groupby(["local_segment"])['inference'].count()/60)/ conv_totalDuration
|
|
|
|
if "silenceexpectedfraction" in features_to_compute:
|
|
conversation_features["silenceexpectedfraction"] = (conversation_data[conversation_data['inference']==0].groupby(["local_segment"])['inference'].count()/60)/ expectedMinutes
|
|
|
|
if "noiseexpectedfraction" in features_to_compute:
|
|
conversation_features["noiseexpectedfraction"] = (conversation_data[conversation_data['inference']==1].groupby(["local_segment"])['inference'].count()/60)/ expectedMinutes
|
|
|
|
if "voiceexpectedfraction" in features_to_compute:
|
|
conversation_features["voiceexpectedfraction"] = (conversation_data[conversation_data['inference']==2].groupby(["local_segment"])['inference'].count()/60)/ expectedMinutes
|
|
|
|
if "unknownexpectedfraction" in features_to_compute:
|
|
conversation_features["unknownexpectedfraction"] = (conversation_data[conversation_data['inference']==3].groupby(["local_segment"])['inference'].count()/60)/ expectedMinutes
|
|
|
|
if "sumconversationduration" in features_to_compute:
|
|
conversation_features["sumconversationduration"] = conversation_data.groupby(["local_segment"])["conv_duration"].sum()
|
|
|
|
if "avgconversationduration" in features_to_compute:
|
|
conversation_features["avgconversationduration"] = conversation_data[conversation_data["conv_duration"] > 0].groupby(["local_segment"])["conv_duration"].mean()
|
|
|
|
if "sdconversationduration" in features_to_compute:
|
|
conversation_features["sdconversationduration"] = conversation_data[conversation_data["conv_duration"] > 0].groupby(["local_segment"])["conv_duration"].std()
|
|
|
|
if "minconversationduration" in features_to_compute:
|
|
conversation_features["minconversationduration"] = conversation_data[conversation_data["conv_duration"] > 0].groupby(["local_segment"])["conv_duration"].min()
|
|
|
|
if "maxconversationduration" in features_to_compute:
|
|
conversation_features["maxconversationduration"] = conversation_data.groupby(["local_segment"])["conv_duration"].max()
|
|
|
|
if "timefirstconversation" in features_to_compute:
|
|
timestampsLastConversation = conversation_data[conversation_data["double_convo_start"] > 0].groupby(["local_segment"])['timestamp'].min()
|
|
if len(list(timestampsLastConversation.index)) > 0:
|
|
for date in list(timestampsLastConversation.index):
|
|
lastimestamp = timestampsLastConversation.loc[date]
|
|
lasttime = (conversation_data.query('timestamp == @lastimestamp', inplace = False))['local_time'].iat[0]
|
|
conversation_features.loc[date,"timefirstconversation"] = int(lasttime.split(':')[0])*60 + int(lasttime.split(':')[1])
|
|
else:
|
|
conversation_features["timefirstconversation"] = np.nan
|
|
|
|
if "timelastconversation" in features_to_compute:
|
|
timestampsLastConversation = conversation_data[conversation_data["double_convo_start"] > 0].groupby(["local_segment"])['timestamp'].max()
|
|
if len(list(timestampsLastConversation.index)) > 0:
|
|
for date in list(timestampsLastConversation.index):
|
|
lastimestamp = timestampsLastConversation.loc[date]
|
|
lasttime = (conversation_data.query('timestamp == @lastimestamp', inplace = False))['local_time'].iat[0]
|
|
conversation_features.loc[date,"timelastconversation"] = int(lasttime.split(':')[0])*60 + int(lasttime.split(':')[1])
|
|
else:
|
|
conversation_features["timelastconversation"] = np.nan
|
|
|
|
if "noisesumenergy" in features_to_compute:
|
|
conversation_features["noisesumenergy"] = conversation_data[conversation_data['inference']==1].groupby(["local_segment"])["double_energy"].sum()
|
|
|
|
if "noiseavgenergy" in features_to_compute:
|
|
conversation_features["noiseavgenergy"] = conversation_data[conversation_data['inference']==1].groupby(["local_segment"])["double_energy"].mean()
|
|
|
|
if "noisesdenergy" in features_to_compute:
|
|
conversation_features["noisesdenergy"] = conversation_data[conversation_data['inference']==1].groupby(["local_segment"])["double_energy"].std()
|
|
|
|
if "noiseminenergy" in features_to_compute:
|
|
conversation_features["noiseminenergy"] = conversation_data[conversation_data['inference']==1].groupby(["local_segment"])["double_energy"].min()
|
|
|
|
if "noisemaxenergy" in features_to_compute:
|
|
conversation_features["noisemaxenergy"] = conversation_data[conversation_data['inference']==1].groupby(["local_segment"])["double_energy"].max()
|
|
|
|
if "voicesumenergy" in features_to_compute:
|
|
conversation_features["voicesumenergy"] = conversation_data[conversation_data['inference']==2].groupby(["local_segment"])["double_energy"].sum()
|
|
|
|
if "voiceavgenergy" in features_to_compute:
|
|
conversation_features["voiceavgenergy"] = conversation_data[conversation_data['inference']==2].groupby(["local_segment"])["double_energy"].mean()
|
|
|
|
if "voicesdenergy" in features_to_compute:
|
|
conversation_features["voicesdenergy"] = conversation_data[conversation_data['inference']==2].groupby(["local_segment"])["double_energy"].std()
|
|
|
|
if "voiceminenergy" in features_to_compute:
|
|
conversation_features["voiceminenergy"] = conversation_data[conversation_data['inference']==2].groupby(["local_segment"])["double_energy"].min()
|
|
|
|
if "voicemaxenergy" in features_to_compute:
|
|
conversation_features["voicemaxenergy"] = conversation_data[conversation_data['inference']==2].groupby(["local_segment"])["double_energy"].max()
|
|
|
|
conversation_features.fillna(value={feature_name: 0 for feature_name in conversation_features.columns if feature_name not in ["timefirstconversation", "timelastconversation", "sdconversationduration", "noisesdenergy", "voicesdenergy"]}, inplace=True)
|
|
conversation_features = conversation_features.reset_index()
|
|
|
|
return conversation_features |