func main() {
//start := time.Now()
rand.Seed(1)
//fmt.Println("Loading query")
flagCpu := runtime.NumCPU()
fragmentLib := util.Library(json)
pdbQueries := make([]string, 1)
pdbQueries[0] = pdbQuery
bows := util.ProcessBowers(pdbQueries, fragmentLib, false, flagCpu, util.FlagQuiet)
// for b := range bows {
// searchQuery.Add(b)
// }
db_centers, _ := bowdb.Open(fragmentLibraryLoc)
db_centers.ReadAll()
//fmt.Println(fmt.Sprintf("\t%d",timer()))
//fmt.Println("Unserializing gob")
db_slices := dec_gob_ss_db(gobLoc)
var m map[string]int
m = make(map[string]int)
for i, center := range db_centers.Entries {
m[center.Id] = i
}
//fmt.Println(fmt.Sprintf("\t%d",timer()))
sortBy := bowdb.SortByEuclid
if metric == cosineDist {
sortBy = bowdb.SortByCosine
}
var coarse_search = bowdb.SearchOptions{
Limit: -1,
Min: 0.0,
Max: (float64(clusterRadius) + float64(maxRadius)),
SortBy: sortBy,
Order: bowdb.OrderAsc,
}
//var fine_search = bowdb.SearchOptions{
//Limit: -1,
//Min: 0.0,
//Max: float64(maxRadius),
//SortBy: bowdb.SortByEuclid,
// Order: bowdb.OrderAsc,
//}
//fmt.Println("Computing coarse results")
for b := range bows {
var coarse_results []bowdb.SearchResult
coarse_results = db_centers.Search(coarse_search, b)
//coarse_results_time := timer()
//fmt.Println(fmt.Sprintf("\t%d",coarse_results_time))
//fmt.Println(fmt.Sprintf("\tCount: %d",len(coarse_results)))
// fmt.Println("Computing fine results")
var fine_results []bowdb.SearchResult
for _, center := range coarse_results {
for _, entry := range db_slices[m[center.Id]] {
var dist float64
switch metric {
case cosineDist:
dist = b.Bow.Cosine(entry.Bow)
case euclideanDist:
dist = b.Bow.Euclid(entry.Bow)
}
if dist <= float64(maxRadius) {
result := newSearchResult(b, entry)
fmt.Printf(entry.Id)
fmt.Printf(" ")
fmt.Printf("%v", dist)
fmt.Printf(" ")
fine_results = append(fine_results, result)
}
}
}
}
}
func main() {
timer()
runtime.GOMAXPROCS(20)
db, _ := bowdb.Open(fragmentLibraryLoc)
db.ReadAll()
//Assert(err, "Could not open BOW database '%s'", path)
var kCenters []bow.Bowed
fmt.Println(fmt.Sprintf("%d: Generating cluster centers", timer()))
if maxRadius > 0 {
kCenters = maxRadiusKCenter(db.Entries, metric, maxRadius)
numCenters = len(kCenters)
} else if centerType == randomSelec {
kCenters = randomKCenter(db.Entries, metric, numCenters)
} else if centerType == metricApprox {
var start_centers []bow.Bowed
kCenters = metricKCenter(db.Entries, metric, numCenters, start_centers)
} else if centerType == halfhalf {
start_centers := randomKCenter(db.Entries, metric, numCenters/2)
kCenters = metricKCenter(db.Entries, metric, numCenters-numCenters/2, start_centers)
}
// for i, center := range kCenters {
// fmt.Println(center.Id + fmt.Sprintf(": %d",i))
// }
runtime.GOMAXPROCS(20)
fmt.Println(fmt.Sprintf("%d: Computing distances from cluster centers", timer()))
db_codes := make([]int, len(db.Entries))
distances := make([]float64, len(db.Entries))
sem := make(chan empty, len(db.Entries))
for j, _ := range db.Entries {
go func(j int) {
distances[j], _, db_codes[j] = distanceFromSet(metric, db.Entries[j], kCenters)
//fmt.Println(strconv.Itoa(i) + " " + strconv.Itoa(j) + " " + strconv.FormatFloat(dist,'f',5,32))
}(j)
sem <- empty{}
}
for i := 0; i < len(db.Entries); i++ {
<-sem
}
runtime.GOMAXPROCS(20)
fmt.Println(fmt.Sprintf("%d: Writing out centers.cluster.db", timer()))
db_centers, _ := bowdb.Create(db.Lib, "centers.cluster.db")
for _, center := range kCenters {
db_centers.Add(center)
}
db_centers.Close()
fmt.Println(fmt.Sprintf("%d: Opening centers library", timer()))
db_centers2, _ := bowdb.Open("centers.cluster.db")
db_centers2.ReadAll()
var mr map[string]int
mr = make(map[string]int)
for i, center := range db_centers2.Entries {
mr[center.Id] = i
}
db_slices := make([][]bow.Bowed, numCenters, numCenters)
fmt.Println(fmt.Sprintf("%d: Computing individual cluster dbs", timer()))
for i := 0; i < len(kCenters); i++ {
//curr_cluster, _ := bowdb.Create(db.Lib, kCenters[i].Id + ".cluster.db")
//curr_cluster := db_slices[mr[kCenters[i].Id]]
for j, entry := range db.Entries {
if i == db_codes[j] {
db_slices[mr[kCenters[i].Id]] = append(db_slices[mr[kCenters[i].Id]], entry)
}
}
//curr_cluster.Close()
}
gobLoc := "clusters.gob"
fmt.Println(fmt.Sprintf("%d: Serializing gob", timer()))
enc_gob_ss_db(db_slices, gobLoc)
fmt.Println(fmt.Sprintf("%d: computing cluster radii", timer()))
cluster_radii := make([]float64, numCenters)
cluster_count := make([]int, numCenters)
for j, _ := range db.Entries {
if distances[j] > cluster_radii[db_codes[j]] {
cluster_radii[db_codes[j]] = distances[j]
}
cluster_count[db_codes[j]]++
}
for j, entry := range kCenters {
fmt.Println(entry.Id + fmt.Sprintf("\t%f\t%d", cluster_radii[j], cluster_count[j]))
}
fmt.Println(fmt.Sprintf("%d: Finished!!", timer()))
}
func OpenBowDB(path string) *bowdb.DB {
db, err := bowdb.Open(path)
Assert(err, "Could not open BOW database '%s'", path)
return db
}
func main() {
rand.Seed(1)
db_query, _ := bowdb.Open(searchQuery)
db_query.ReadAll()
var query bow.Bowed
query = db_query.Entries[0]
db_centers, _ := bowdb.Open(fragmentLibraryLoc)
db_centers.ReadAll()
db_slices := dec_gob_ss_db("clusters.gob")
var m map[string]int
m = make(map[string]int)
for i, center := range db_centers.Entries {
m[center.Id] = i
}
/*
sortBy := bowdb.SortByEuclid
if metric == cosineDist {
sortBy = bowdb.SortByCosine
}
*/
db, _ := bowdb.Open(potentialTargetsLoc)
db.ReadAll()
//repeatNum := 10 // How many times to repeat each run for timing purposes
fmt.Println("Radius\tAccelCount\tLongCount\tAccel\tNaive\tSpeedup\tSensitivity\tFineCandidates")
for maxR := 0; maxR < 50; maxR = maxR + 1 {
maxRadius := 0.0
if metric == cosineDist {
maxRadius = float64(maxR) / 100.0
} else {
maxRadius = float64(maxR)
}
coarse_radius := float64(clusterRadius) + float64(maxRadius)
accelCount := make([]int, repeatNum)
longCount := make([]int, repeatNum)
accelTime := make([]int64, repeatNum)
naiveTime := make([]int64, repeatNum)
fineCandidates := make([]int, repeatNum)
for rep := 0; rep < repeatNum; rep++ {
timer()
var coarse_results []bowdb.SearchResult
//coarse_results = db_centers.Search(coarse_search, query)
for _, entry := range db_centers.Entries {
var dist float64
switch metric {
case cosineDist:
dist = query.Bow.Cosine(entry.Bow)
case euclideanDist:
dist = query.Bow.Euclid(entry.Bow)
}
if dist <= coarse_radius {
result := newSearchResult(query, entry)
coarse_results = append(coarse_results, result)
}
}
coarse_results_time := timer()
var fine_results []bowdb.SearchResult
fine_candidates := 0
for _, center := range coarse_results {
fine_candidates += len(db_slices[m[center.Id]])
for _, entry := range db_slices[m[center.Id]] {
var dist float64
switch metric {
case cosineDist:
dist = query.Bow.Cosine(entry.Bow)
case euclideanDist:
dist = query.Bow.Euclid(entry.Bow)
}
if dist <= float64(maxRadius) {
result := newSearchResult(query, entry)
fine_results = append(fine_results, result)
}
}
}
fine_results_time := timer()
var long_results []bowdb.SearchResult
//long_results = db.Search(fine_search, query)
for _, entry := range db.Entries {
var dist float64
switch metric {
case cosineDist:
dist = query.Bow.Cosine(entry.Bow)
case euclideanDist:
dist = query.Bow.Euclid(entry.Bow)
}
if dist <= float64(maxRadius) {
result := newSearchResult(query, entry)
long_results = append(long_results, result)
}
}
long_results_time := timer()
/*if (len(long_results)!=len(fine_results)) {
err := "Fine and long searches did not match."
fmt.Fprintf(os.Stderr, "error: %v\n", err)
fmt.Fprintf(os.Stderr, "Fine: %v\n", len(fine_results))
fmt.Fprintf(os.Stderr, "Long: %v\n", len(long_results))
os.Exit(1)
} */
long_count := len(long_results)
fine_count := len(fine_results)
accel_time := coarse_results_time + fine_results_time
accelCount[rep] = fine_count
longCount[rep] = long_count
accelTime[rep] = accel_time
naiveTime[rep] = long_results_time
fineCandidates[rep] = fine_candidates
}
accelCountAvg := averageInt2F64(accelCount)
naiveCountAvg := averageInt2F64(longCount)
accelTimeAvg := averageInt642F64(accelTime)
naiveTimeAvg := averageInt642F64(naiveTime)
sensitivity := accelCountAvg / naiveCountAvg
speedup := naiveTimeAvg / accelTimeAvg
fineSearchCount := averageInt2F64(fineCandidates)
fmt.Println(fmt.Sprintf("%f\t%f\t%f\t%f\t%f\t%f\t%f\t%f", maxRadius, accelCountAvg, naiveCountAvg, accelTimeAvg, naiveTimeAvg, speedup, sensitivity, fineSearchCount))
}
}
