use stats to determine how much to extract

killer app command for the moment

dupi index <stuff>
dupi extract | awk '{print $1}' | xargs dupi unblot

added -all to unblot

Author: scott-cotton

blot.go

@@ -44,8 +44,12 @@ func (b *Blot) Len() int {
 	return len(b.Docs)
 }
 
-func (b *Blot) Next() *Doc {
+func (b *Blot) Next(lim bool) *Doc {
 	n := len(b.Docs)
-	b.Docs = b.Docs[:n+1]
+	if lim {
+		b.Docs = b.Docs[:n+1]
+	} else {
+		b.Docs = append(b.Docs, Doc{})
+	}
 	return &b.Docs[n]
 }

cmd/dupi/extract.go

@@ -19,6 +19,8 @@ import (
 	"flag"
 	"fmt"
 	"io"
+	"log"
+	"math"
 	"os"
 
 	"github.com/go-air/dupi"
@@ -28,6 +30,7 @@ type extractCmd struct {
 	subCmd
 	index *dupi.Index
 	json  *bool
+	sigma *float64
 }
 
 func newExtractCmd() *extractCmd {
@@ -37,6 +40,7 @@ func newExtractCmd() *extractCmd {
 		flags: flag.NewFlagSet("extract", flag.ExitOnError)}}
 
 	extract.json = extract.flags.Bool("json", false, "output json")
+	extract.sigma = extract.flags.Float64("sigma", 2.0, "explore blots within σ of average (higher=most probable dups, lower=more volume)")
 	return extract
 }
 
@@ -52,10 +56,14 @@ func (x *extractCmd) Run(args []string) error {
 		return err
 	}
 	defer x.index.Close()
+	st, err := x.index.Stats()
+	if err != nil {
+		log.Fatal(err)
+	}
+	σ := *x.sigma
+	N := int(math.Round(st.BlotMean + σ*st.BlotSigma))
 	query := x.index.StartQuery(dupi.QueryMaxBlot)
-	shape := []dupi.Blot{
-		{Blot: 0, Docs: make([]dupi.Doc, 0, 32)},
-		{Blot: 0, Docs: make([]dupi.Doc, 0, 32)}}
+	shape := []dupi.Blot{{Blot: 0}}
 	for {
 		n, err := query.Next(shape)
 		if err == io.EOF {
@@ -67,6 +75,9 @@ func (x *extractCmd) Run(args []string) error {
 		if n == 0 {
 			return fmt.Errorf("Query.Next gave 0 and no error")
 		}
+		if len(shape[0].Docs) < N {
+			return nil
+		}
 		if *x.json {
 			shp2 := shape
 			j := 0
@@ -98,7 +109,7 @@ func (x *extractCmd) Run(args []string) error {
 			}
 		}
 		for i := range shape {
-			shape[i].Docs = shape[i].Docs[:0]
+			shape[i].Docs = nil
 		}
 	}
 }

cmd/dupi/index.go

@@ -45,6 +45,7 @@ func newIndexCmd() *indexCmd {
 	index.add = index.flags.Bool("a", false, "add to a given existing index")
 	index.verbose = index.flags.Bool("v", false, "verbose")
 	index.nshat = index.flags.Int("s", 4, "num shatterers")
+	index.shards = index.flags.Int("n", 4, "num shards")
 	return index
 }
 

cmd/dupi/unblot.go

@@ -10,11 +10,14 @@ import (
 
 type unblotCmd struct {
 	subCmd
+	all *bool
 }
 
 func newUnblotCmd() *unblotCmd {
-	return &unblotCmd{
+	cmd := &unblotCmd{
 		subCmd: subCmd{name: "unblot", flags: flag.NewFlagSet("unblot", flag.ExitOnError)}}
+	cmd.all = cmd.flags.Bool("all", false, "output all matches")
+	return cmd
 }
 
 func (ub *unblotCmd) Usage() string {
@@ -54,7 +57,7 @@ func (ub *unblotCmd) Run(args []string) error {
 			m[dat] = append(m[dat], doc)
 		}
 		for k, ds := range m {
-			if len(ds) < 2 {
+			if !*ub.all && len(ds) < 2 {
 				continue
 			}
 			fmt.Printf("text:\n'''\n%s'''\n", k)

query.go

@@ -67,11 +67,7 @@ func (q *Query) Get(blot *Blot) error {
 		if err != nil {
 			return err
 		}
-		if !lim && len(blot.Docs) == cap(blot.Docs) {
-			blot.Docs = append(blot.Docs, Doc{})
-			blot.Docs = blot.Docs[:len(blot.Docs)-1]
-		}
-		if err = q.index.docid2Doc(docid, blot.Next()); err != nil {
+		if err = q.index.docid2Doc(docid, blot.Next(lim)); err != nil {
 			return err
 		}
 	}
@@ -97,13 +93,18 @@ func (q *Query) Next(dst []Blot) (n int, err error) {
 			}
 			continue
 		}
+		lim := dstBlot.Docs != nil
 		_, err = q.fillBlot(dstBlot, shardState, state.i)
 		if err != nil {
 			return
 		}
 		if len(dstBlot.Docs) <= 1 {
 			q.advance(shardState, state.i)
-			dstBlot.Docs = dstBlot.Docs[:0]
+			if lim {
+				dstBlot.Docs = dstBlot.Docs[:0]
+			} else {
+				dstBlot.Docs = nil
+			}
 			continue
 		}
 		n++
@@ -116,20 +117,20 @@ func (q *Query) fillBlot(dst *Blot, src *shard.ReadState, srcPos uint32) (int, e
 		docid uint32
 		err   error
 		n     int
+		lim   bool
 	)
 	dst.Blot = uint32(src.Blot)*q.state.n + q.state.i
-	for dst.Len() < dst.Cap() {
+	lim = dst.Docs != nil
+	for !lim || dst.Len() < dst.Cap() {
 		docid, err = src.Next()
 		if err == io.EOF {
 			q.advance(src, srcPos)
 			return n, nil
 		} else if err != nil {
 			return 0, err
-		} else if docid == 0 {
-			continue
 		}
 		n++
-		q.index.docid2Doc(docid, dst.Next())
+		q.index.docid2Doc(docid, dst.Next(lim))
 	}
 	return n, err
 }