-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathuseGMMFitToOps.R
219 lines (144 loc) · 10 KB
/
useGMMFitToOps.R
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
209
210
211
212
213
214
215
216
217
218
## 6.22.10 Use GMMs that I fit to the op data to try predicting the query latency
rm(list=ls())
source("/Users/radlab/Desktop/ksauer/Desktop/scads/experiments/client/performance/logparsing/src/main/R/emFitAndSample.R")
load("~/Desktop/ops.RData")
k=5
numRdmInits=5
pathPrefix = "~/Desktop/op3-gettingParams"
for (i in 1:numRdmInits) {
print(paste(("rdm init", i)))
emParams = fitEM(observationsVector=op3$latency_ms, k=k, distrType="gaussian", modelName="V", saveRdmInit=TRUE, destinationPath= pathPrefix, initNum=i)
}
# Load in em results; pick the one with the best loglik
emLogLik = vector(mode="numeric", length=numRdmInits)
for (i in 1:numRdmInits) {
load(paste(pathPrefix, "/rdmInitAndParams-k=", k, "-init", i, ".RData", sep=""))
emLogLik[i] = emRun$loglik
}
bestRdmInit = which.max(emLogLik) # => best rdm init
emLogLik
bestRdmInit
# Use params from best rdm init
load(paste(pathPrefix, "/rdmInitAndParams-k=", k, "-init", bestRdmInit, ".RData", sep=""))
emParams = emRun$parameters
op3GMMSamples = sampleFromMixtureModel(numSamples=1000, distrType="gaussian", parameters=emParams, truncateLessThanZero=TRUE)
# Turn the above param acquisition into a function
getGMMParamsGivenK = function(observationsVector, pathPrefix, k, numRdmInits=5) {
source("/Users/radlab/Desktop/ksauer/Desktop/scads/experiments/client/performance/logparsing/src/main/R/emFitAndSample.R")
# Save the params obtained from several random inits
for (i in 1:numRdmInits) {
print(paste("rdm init", i))
emParams = fitEM(observationsVector=observationsVector, k=k, distrType="gaussian", modelName="V", saveRdmInit=TRUE, destinationPath=pathPrefix, initNum=i)
}
# Determine which random init yielded the best param vtr
emLogLik = vector(mode="numeric", length=numRdmInits)
for (i in 1:numRdmInits) {
load(paste(pathPrefix, "/rdmInitAndParams-k=", k, "-init", i, ".RData", sep=""))
emLogLik[i] = emRun$loglik
}
bestRdmInit = which.max(emLogLik) # => best rdm init
# Return the best params
load(paste(pathPrefix, "/rdmInitAndParams-k=", k, "-init", bestRdmInit, ".RData", sep=""))
emParams = emRun$parameters
return(emParams)
}
getBestGMMParamVector = function(pathPrefix, k, numRdmInits=5) {
# Determine which random init yielded the best param vtr
emLogLik = vector(mode="numeric", length=numRdmInits)
for (i in 1:numRdmInits) {
load(paste(pathPrefix, "/rdmInitAndParams-k=", k, "-init", i, ".RData", sep=""))
emLogLik[i] = emRun$loglik
}
bestRdmInit = which.max(emLogLik) # => best rdm init
# Return the best params
load(paste(pathPrefix, "/rdmInitAndParams-k=", k, "-init", bestRdmInit, ".RData", sep=""))
emParams = emRun$parameters
return(emParams)
}
# try plugging into query latency prediction
# function is in analyticalModelingMassAndTailSeparately.R
checkSingleOpModelImpactOnQueryModel(opSamples=op3GMMSamples, opNum=3, prefix="~/Desktop/checking-gmm-on-query-distr-truncate", queryType="userByHometown")
checkSingleOpModelImpactOnQueryModel(opSamples=op3GMMSamples, opNum=3, prefix="~/Desktop/checking-gmm-on-query-distr-truncate", queryType="needsApproval")
# Check log likelihood computation
#myLogLik = computeGMMLogLikelihood(op3$latency_ms, emParams)
#myLogLik
#kFoldCVToChooseNumMixtureComponentsForGMM(dataVector=op3$latency_ms, destinationPath="~/Desktop/op3-cv-loglik")
# Check out several fits for each op
checkSeveralFits = function(opData, opNum, fits=c("gamma", "lognormal", "gmm"), prefix, gmmParams) {
dir.create(prefix)
source("/Users/radlab/Desktop/ksauer/Desktop/scads/experiments/client/performance/logparsing/src/main/R/graphingAssistFunctions.R")
source("/Users/radlab/Desktop/ksauer/Desktop/scads/experiments/client/performance/logparsing/src/main/R/getDistrSamplers.R")
source("/Users/radlab/Desktop/ksauer/Desktop/scads/experiments/client/performance/logparsing/src/main/R/emFitAndSample.R")
pdf(paste(prefix, "/checkFits-op", opNum, ".pdf", sep=""))
par(mfrow=c(length(fits)+1,1), mar=c(5,5,4,2))
hist(opData, breaks=100)
plotMed90th99thQuantilesAndLegend(opData)
for (i in 1:length(fits)) {
if (fits[i] == "gamma") {
samples = getGammaSamples(observationsVector=opData, numSamples=1000)
} else if (fits[i] == "lognormal") {
samples = getLogNormalSamples(observationsVector=opData, numSamples=1000)
} else if (fits[i] == "gmm") {
samples = sampleFromMixtureModel(numSamples=1000, distrType="gaussian", parameters=gmmParams, truncateLessThanZero=TRUE)
} else {
"Unsupported fit type."
}
hist(samples, breaks=100)
plotMed90th99thQuantilesAndLegend(samples)
}
dev.off()
}
rm(list=ls())
load("~/Desktop/ops.RData")
load("/Users/radlab/Desktop/op3-gettingParams/rdmInitAndParams-k=5-init5.RData")
ls()
checkSeveralFits(op3$latency_ms, opNum=3, prefix="~/Desktop/fits", gmmParams=emRun$parameters)
checkSeveralFits(op3$latency_ms, opNum=3, prefix="~/Desktop/fits2", gmmParams=getGMMParamsGivenK(observationsVector=op3$latency_ms, pathPrefix="~/Desktop/gmmParamFcn", k=5))
op = op9
opNum = 9
k=c(3,8,5,3,5,4,4,5,2)
checkSeveralFits(op$latency_ms, opNum=opNum, prefix="~/Desktop/fits", gmmParams=getGMMParamsGivenK(observationsVector=op$latency_ms, pathPrefix=paste("~/Desktop/op", opNum, "GMMParams", sep=""), k=k[opNum]), fits="gmm")
# Get ops' GMM samples and use to do query latency pred
rm(list=ls())
source("/Users/radlab/Desktop/ksauer/Desktop/scads/experiments/client/performance/logparsing/src/main/R/emFitAndSample.R")
load("~/Desktop/ops.RData")
k=c(3,8,5,3,5,4,4,5,2) # from CV
op1 = sampleFromMixtureModel(numSamples=1000, distrType="gaussian", parameters=getGMMParamsGivenK(observationsVector=op1$latency_ms, pathPrefix="~/Desktop/gmmParamFcn", k=k[1]), truncateLessThanZero=TRUE)
op2 = sampleFromMixtureModel(numSamples=1000, distrType="gaussian", parameters=getGMMParamsGivenK(observationsVector=op2$latency_ms, pathPrefix="~/Desktop/gmmParamFcn", k=k[2]), truncateLessThanZero=TRUE)
op3 = sampleFromMixtureModel(numSamples=1000, distrType="gaussian", parameters=getGMMParamsGivenK(observationsVector=op3$latency_ms, pathPrefix="~/Desktop/gmmParamFcn", k=k[3]), truncateLessThanZero=TRUE)
op4 = sampleFromMixtureModel(numSamples=1000, distrType="gaussian", parameters=getGMMParamsGivenK(observationsVector=op4$latency_ms, pathPrefix="~/Desktop/gmmParamFcn", k=k[4]), truncateLessThanZero=TRUE)
op5 = sampleFromMixtureModel(numSamples=1000, distrType="gaussian", parameters=getGMMParamsGivenK(observationsVector=op5$latency_ms, pathPrefix="~/Desktop/gmmParamFcn", k=k[5]), truncateLessThanZero=TRUE)
op6 = sampleFromMixtureModel(numSamples=1000, distrType="gaussian", parameters=getGMMParamsGivenK(observationsVector=op6$latency_ms, pathPrefix="~/Desktop/gmmParamFcn", k=k[6]), truncateLessThanZero=TRUE)
op7 = sampleFromMixtureModel(numSamples=1000, distrType="gaussian", parameters=getGMMParamsGivenK(observationsVector=op7$latency_ms, pathPrefix="~/Desktop/gmmParamFcn", k=k[7]), truncateLessThanZero=TRUE)
op8 = sampleFromMixtureModel(numSamples=1000, distrType="gaussian", parameters=getGMMParamsGivenK(observationsVector=op8$latency_ms, pathPrefix="~/Desktop/gmmParamFcn", k=k[8]), truncateLessThanZero=TRUE)
op9 = sampleFromMixtureModel(numSamples=1000, distrType="gaussian", parameters=getGMMParamsGivenK(observationsVector=op9$latency_ms, pathPrefix="~/Desktop/gmmParamFcn", k=k[9]), truncateLessThanZero=TRUE)
save(op1, op2, op3, op4, op5, op6, op7, op8, op9, file="~/Desktop/opsGMM.RData")
# Oops; should save hist, not samples
op1distr = hist(op1distr, breaks=1000)
op2distr = hist(op2distr, breaks=1000)
op3distr = hist(op3distr, breaks=1000)
op4distr = hist(op4distr, breaks=1000)
op5distr = hist(op5distr, breaks=1000)
op6distr = hist(op6distr, breaks=1000)
op7distr = hist(op7distr, breaks=1000)
op8distr = hist(op8distr, breaks=1000)
op9distr = hist(op9distr, breaks=1000)
save(op1distr, op2distr, op3distr, op4distr, op5distr, op6distr, op7distr, op8distr, op9distr, file="~/Desktop/gmmOpDistr.RData")
# Checking query pred
source("/Users/radlab/Desktop/ksauer/Desktop/scads/experiments/client/performance/logparsing/src/main/R/modeling-harness-analytical-distr.R")
computeAndComparePredictedQueryDistr(queryObsFile="~/Desktop/queryObs.RData", opDistrFile="~/Desktop/gmmOpDistr.RData", destinationPath="~/Desktop/gmmQueryPred-1kSamples", queryType="userByHometown", numSamples=1000)
computeAndComparePredictedQueryDistr(queryObsFile="~/Desktop/queryObs.RData", opDistrFile="~/Desktop/gmmOpDistr.RData", destinationPath="~/Desktop/gmmQueryPred-1kSamples", queryType="needsApproval", numSamples=1000)
computeAndComparePredictedQueryDistr(queryObsFile="~/Desktop/queryObs.RData", opDistrFile="~/Desktop/gmmOpDistr.RData", destinationPath="~/Desktop/gmmQueryPred-1kSamples", queryType="myFollowing", numSamples=1000)
computeAndComparePredictedQueryDistr(queryObsFile="~/Desktop/queryObs.RData", opDistrFile="~/Desktop/gmmOpDistr.RData", destinationPath="~/Desktop/gmmQueryPred-1kSamples", queryType="myThoughts", numSamples=1000)
# Save params for each op
op1params = getBestGMMParamVector(pathPrefix=paste("~/Desktop/op1GMMParams", sep=""), k=k[1])
op2params = getBestGMMParamVector(pathPrefix=paste("~/Desktop/op2GMMParams", sep=""), k=k[2])
op3params = getBestGMMParamVector(pathPrefix=paste("~/Desktop/op3GMMParams", sep=""), k=k[3])
op4params = getBestGMMParamVector(pathPrefix=paste("~/Desktop/op4GMMParams", sep=""), k=k[4])
op5params = getBestGMMParamVector(pathPrefix=paste("~/Desktop/op5GMMParams", sep=""), k=k[5])
op6params = getBestGMMParamVector(pathPrefix=paste("~/Desktop/op6GMMParams", sep=""), k=k[6])
op7params = getBestGMMParamVector(pathPrefix=paste("~/Desktop/op7GMMParams", sep=""), k=k[7])
op8params = getBestGMMParamVector(pathPrefix=paste("~/Desktop/op8GMMParams", sep=""), k=k[8])
op9params = getBestGMMParamVector(pathPrefix=paste("~/Desktop/op9GMMParams", sep=""), k=k[9])
save(op1params, op2params, op3params, op4params, op5params, op6params, op7params, op8params, op9params, file="~/Desktop/opGMMParams.RData")
computeAndComparePredictedQueryDistr(queryObsFile="~/Desktop/queryObs.RData", opDistrFile="~/Desktop/opGMMParams.RData", destinationPath="~/Desktop/gmmQueryPred-gmmSampler2", queryType="userByHometown", numSamples=10000)