textsimilarity/density_graphs.R at master · meedan/textsimilarity · GitHub

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
library(ggplot2)
library(plyr)

df<-read.csv('output_measures_textsim_norm.csv',stringsAsFactors = FALSE)
df2<-read.csv("output_multilingual_unisent.csv",sep="\t",stringsAsFactors = FALSE)
df3<-read.csv("output_flair.csv",sep="\t",stringsAsFactors = FALSE)
dfData<-rbind(df,df2,df3)
head(dfData)

p<-ggplot(dfData,aes(x=metric,y=measure,fill=which_comparisons))+geom_boxplot()+coord_flip()
p<-p+theme_bw()+theme(legend.position="bottom",legend.title=element_blank())
ggsave("boxplots.png",p,width=10,height=8)


p<-ggplot(dfData,aes(x=measure,fill=which_comparisons))+geom_density(alpha=0.5)
p<-p+facet_wrap(~metric)+theme_bw()+theme(legend.position="bottom",legend.title=element_blank())
p

ggsave("density_plots.png",p,width=12,height=8)


unique(dfData$metric)

dfDataSub<-subset(dfData,metric%in%c("set","gn300model-cos","gn300model-word-mover","multi-unisent-angdist","multi-unisent-cosine","web-bert"))

dfDataSub$metric<-factor(dfDataSub$metric,levels=c("set","gn300model-cos","gn300model-word-mover","multi-unisent-cosine","multi-unisent-angdist","web-bert"))

p<-ggplot(dfDataSub,aes(x=measure,fill=which_comparisons))+geom_density(alpha=0.5)
p<-p+facet_wrap(~metric)+theme_bw()+theme(legend.position="bottom",legend.title=element_blank())
p

ggsave("density_plots_subset.png",p,width=12,height=8)


##############
# Compare some thresholds

thresholds<-c(1,.99,.97,.95,.9,.85,.8)
metrics<-c("gn300model-cos","gn300model-ang","gn300model-sqrt","multi-unisent-angdist","multi-unisent-cosine","es-match", "set", "xlmr-wmdist")#,"CR5")

tsub<-subset(dfData,metric%in%metrics)


for (t in thresholds) {
  #dfData[,paste0("t",t)]<-dfData$measure>t
  print("------------------------")
  print(paste0("Threshold: ",t))

  vals<-tsub$measure>=t
  sub<-tsub[vals,]
  x<-ddply(sub,.(metric),function(df) {
    return(data.frame(
      len=nrow(df),
      true_pos=sum(df$which_comparisons=="SAME")/nrow(df),
      false_pos=sum(df$which_comparisons=="DIFF")/nrow(df)
    ))
  })
  print(x)
}

print("Total 'good' matches")
sum(tsub$which_comparisons=="SAME" & tsub$metric=="gn300model-cos")
print("Total 'bad' matches")
sum(tsub$which_comparisons=="DIFF" & tsub$metric=="gn300model-cos")
print("Total sentence pairs in data")
sum(tsub$metric=="gn300model-cos")

########################
library(ggplot2)
library(plyr)

dfHindi<-read.csv('output_hindi.csv',sep="\t",stringsAsFactors = FALSE)

p<-ggplot(dfHindi,aes(x=metric,y=measure,fill=which_comparisons))+geom_boxplot()+coord_flip()
p<-p+theme_bw()+theme(legend.position="bottom",legend.title=element_blank())
ggsave("hindi_boxplots.png",p,width=10,height=8)


p<-ggplot(dfHindi,aes(x=measure,fill=which_comparisons))+geom_density(alpha=0.5)
p<-p+facet_wrap(~metric,scales="free")+theme_bw()+theme(legend.position="bottom",legend.title=element_blank())
p

ggsave("hindi_density_plots.png",p,width=12,height=8)