-
Notifications
You must be signed in to change notification settings - Fork 2
/
Copy pathsentiment_feature_generator.py
208 lines (182 loc) · 10.6 KB
/
sentiment_feature_generator.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
import pandas as pd
import numpy as np
from datetime import datetime
import sys
from functools import reduce
from collections import Counter
# when debug mode is on, we only take a sub-sample of total data
debug_mode = True
# when we first load this in notebook, turn reload on. afterwards, turn it off no need to reload data everytime
reload = True
if reload:
news_train_dir = "./new_train_df.csv"
news_train_df = pd.read_csv(news_train_dir)
market_train_dir = "./market_train_df.csv"
market_train_df = pd.read_csv(market_train_dir)
# globals
news_col_extractor = ["time", "assetCodes", "headline", "urgency", "takeSequence",
"subjects", "audiences", "relevance",
'sentimentClass','sentimentNegative', 'sentimentNeutral', 'sentimentPositive',
'noveltyCount12H', 'noveltyCount24H', 'noveltyCount3D', 'noveltyCount5D', 'noveltyCount7D',
'volumeCounts12H','volumeCounts24H', 'volumeCounts3D', 'volumeCounts5D','volumeCounts7D'
]
market_col_extractor = ["time", "assetCode", "volume", "close", "open",
"returnsClosePrevRaw1", "returnsOpenPrevRaw1", "returnsClosePrevRaw10", "returnsOpenPrevRaw10",
"returnsOpenNextMktres10", "universe"]
identity = lambda series: reduce(lambda x, y: x, series)
coalesce = lambda x: list(x)
if debug_mode:
news = news_train_df[60000:70000]
market = market_train_df[:5000]
else:
news = news_train_df
market = market_train_df
assetCode_set = set(market_train_df["assetCode"].unique())
# extract relevant columns based on their descriptions
def extract_df(news_train_df, market_train_df):
news_df = news_train_df[news_col_extractor]
market_df = market_train_df[market_col_extractor]
return news_df, market_df
# given a dataframe with field time convert into datetime, month and week
def extract_time_dependent_features(df, obj = None):
# 1. get date
df["datetime"] = df["time"].apply(lambda ts: ts[:10])
# 2. get month
if obj == "news":
return df
df["month"] = df["datetime"].apply(lambda ts: ts[5:7])
# 3. get week
df["week"] = df["datetime"].apply(lambda ts: datetime.strptime(ts, '%Y-%m-%d').strftime('%a'))
return df
# apply helper aggregator to reduce assetCode in preparation for joining
def assetCodeMapper(assetCodeSet):
assets = list(eval(assetCodeSet).intersection(assetCode_set))
if assets == []:
return None
else:
return assets[0]
# join if assetCode in assetCodes and time days are the same
def mergeDframes(news_df, market_df):
anchor = ["datetime", "assetCode"]
mergedDF = market_df.merge(news_df, on=["datetime", "assetCode"], how="left").dropna()
return mergedDF
# squash columns so that (datetime, assetCodes) are unique
def squash(res):
df = res.groupby("datetime")
df = res.groupby(["datetime", "assetCode"]).agg({'volume': identity,
'open': identity,
'close': identity,
'returnsClosePrevRaw1': identity,
'returnsOpenPrevRaw1': identity,
'returnsClosePrevRaw10': identity,
'universe': identity,
'month': identity,
'week': identity,
'headline': coalesce,
'urgency': coalesce,
'takeSequence': coalesce,
'subjects': coalesce,
'audiences': coalesce,
'relevance': coalesce,
'sentimentClass': coalesce,
'sentimentNegative': coalesce,
'sentimentNeutral': coalesce,
'sentimentPositive': coalesce,
'noveltyCount12H': coalesce,
'noveltyCount24H': coalesce,
'noveltyCount3D': coalesce,
'noveltyCount5D': coalesce,
'noveltyCount7D': coalesce,
'volumeCounts12H': coalesce,
'volumeCounts24H': coalesce,
'volumeCounts3D': coalesce,
'volumeCounts5D': coalesce,
'volumeCounts7D': coalesce,
'returnsOpenNextMktres10': identity
})
return df
# helper functiuons for urgency related partition calculation
def urgency_helper(x, column, urgency_type):
relevance = [0 if i == urgency_type else i for i in x.relevance]
return np.multiply(relevance, column).sum() / sum(relevance)
def urgency_dist_helper(x):
d = Counter(x)
if 1 not in d:
d[1] = 0
if 3 not in d:
d[3] = 0
return d[1], d[3]
def urgency_time_helper(x, column, urgency_type):
# as indicator function
relevance = [0 if i == urgency_type else 1 for i in x.relevance]
return np.multiply(relevance, column).sum()
# generate relevance weighted features
def generate_relevance_weighted_sentiment(squashedDf):
# we are removing urgency = 2 type because there are too few of them for learning
squashedDf = squashedDf[squashedDf["urgency"] != 2]
# for article and alert, let's compute different values
urgency_ls = [1, 3]
urgency_name = ["alert", "article"]
time_ls = ["12H", "24H", "3D", "5D", "7D"]
for i in range(len(urgency_ls)):
name = urgency_name[i]
u = urgency_ls[i]
squashedDf[name+"_relevance_weighted_sentiment"] = squashedDf.apply(lambda x: urgency_helper(x, x.sentimentClass, u), axis=1)
squashedDf[name+"_relevance_weighted_negative_sentiment"] = squashedDf.apply(lambda x: urgency_helper(x, x.sentimentNegative, u), axis=1)
squashedDf[name+"_relevance_weighted_positive_sentiment"] = squashedDf.apply(lambda x: urgency_helper(x, x.sentimentPositive, u), axis=1)
squashedDf[name+"_relevance_weighted_neutral_sentiment"] = squashedDf.apply(lambda x: urgency_helper(x, x.sentimentNeutral, u), axis=1)
for time in time_ls:
squashedDf[name+"_news_volume_sum_"+time] = squashedDf.apply(lambda x: urgency_time_helper(x, x["volumeCounts"+time], u), axis=1)
squashedDf[name+"_news_novelty_sum_"+time] = squashedDf.apply(lambda x: urgency_time_helper(x, x["noveltyCount"+time], u), axis=1)
squashedDf["relevance_weighted_sentiment"] = squashedDf.apply(lambda x: np.multiply(x.relevance, x.sentimentClass).sum() / sum(x.relevance), axis=1)
squashedDf["relevance_weighted_negative_sentiment"] = squashedDf.apply(lambda x: np.multiply(x.relevance, x.sentimentNegative).sum() / sum(x.relevance), axis=1)
squashedDf["relevance_weighted_positive_sentiment"] = squashedDf.apply(lambda x: np.multiply(x.relevance, x.sentimentPositive).sum() / sum(x.relevance), axis=1)
squashedDf["relevance_weighted_neutral_sentiment"] = squashedDf.apply(lambda x: np.multiply(x.relevance, x.sentimentNeutral).sum(), axis=1)
for time in time_ls:
squashedDf["news_volume_sum_"+time] = squashedDf.apply(lambda x: sum(x["volumeCounts"+time]), axis=1)
squashedDf["news_novelty_sum_"+time] = squashedDf.apply(lambda x: sum(x["noveltyCount"+time]), axis=1)
squashedDf["alert"] = squashedDf.urgency.apply(lambda x: urgency_dist_helper(x)[0])
squashedDf["article"] = squashedDf.urgency.apply(lambda x: urgency_dist_helper(x)[1])
return squashedDf
def extract_features(df):
extract_ls = ['month','week', 'alert', 'article',
'alert_relevance_weighted_sentiment',
'alert_relevance_weighted_negative_sentiment',
'alert_relevance_weighted_positive_sentiment',
'alert_relevance_weighted_neutral_sentiment',
'alert_news_volume_sum_12H', 'alert_news_novelty_sum_12H',
'alert_news_volume_sum_24H', 'alert_news_novelty_sum_24H',
'alert_news_volume_sum_3D', 'alert_news_novelty_sum_3D',
'alert_news_volume_sum_5D', 'alert_news_novelty_sum_5D',
'alert_news_volume_sum_7D', 'alert_news_novelty_sum_7D',
'article_relevance_weighted_sentiment',
'article_relevance_weighted_negative_sentiment',
'article_relevance_weighted_positive_sentiment',
'article_relevance_weighted_neutral_sentiment',
'article_news_volume_sum_12H', 'article_news_novelty_sum_12H',
'article_news_volume_sum_24H', 'article_news_novelty_sum_24H',
'article_news_volume_sum_3D', 'article_news_novelty_sum_3D',
'article_news_volume_sum_5D', 'article_news_novelty_sum_5D',
'article_news_volume_sum_7D', 'article_news_novelty_sum_7D',
'relevance_weighted_sentiment', 'relevance_weighted_negative_sentiment',
'relevance_weighted_positive_sentiment',
'relevance_weighted_neutral_sentiment', 'news_volume_sum_12H',
'news_novelty_sum_12H', 'news_volume_sum_24H', 'news_novelty_sum_24H',
'news_volume_sum_3D', 'news_novelty_sum_3D', 'news_volume_sum_5D',
'news_novelty_sum_5D', 'news_volume_sum_7D', 'news_novelty_sum_7D',
'returnsOpenNextMktres10']
df = df[df.universe == 1.0][extract_ls].dropna()
return df
# orchestration
def main():
news_df, market_df = extract_df(news, market)
market_df = extract_time_dependent_features(market_df)
news_df = extract_time_dependent_features(news_df, "news")
news_df["assetCode"] = news_df["assetCodes"].apply(assetCodeMapper)
mergedDF = mergeDframes(news_df, market_df)
squashedDf = squash(mergedDF)
featureDf = generate_relevance_weighted_sentiment(squashedDf)
df = extract_features(featureDf)
return df
# execution
df = main()