func main() {
var (
in *fasta.Reader
out, profile *os.File
e error
)
inName := flag.String("in", "", "Filename for input to be factorised. Defaults to stdin.")
outName := flag.String("out", "", "Filename for output. Defaults to stdout.")
k := flag.Int("k", 8, "kmer size to use.")
cat := flag.Int("cat", 5, "number of categories.")
iter := flag.Int("i", 1000, "iterations.")
limit := flag.Int("time", 10, "time limit for NMF.")
lo := flag.Int("lo", 1, "minimum number of kmer frequency to use in NMF.")
hi := flag.Float64("hi", 0.5, "maximum proportion of kmer representation to use in NMF.")
sf := flag.Float64("sf", 0.01, "factor for sparcity of estimating matrices for NMF.")
tol := flag.Float64("tol", 0.001, "tolerance for NMF.")
threads := flag.Int("threads", 2, "number of threads to use.")
seed := flag.Int64("seed", -1, "seed for random number generator (-1 uses system clock).")
cpuprofile := flag.String("cpuprofile", "", "write cpu profile to this file.")
help := flag.Bool("help", false, "print this usage message.")
flag.Parse()
if *help {
flag.Usage()
os.Exit(1)
}
runtime.GOMAXPROCS(*threads)
sparse.MaxProcs = *threads
fmt.Fprintf(os.Stderr, "Using %d threads.\n", runtime.GOMAXPROCS(0))
if *cpuprofile != "" {
if profile, e = os.Create(*cpuprofile); e != nil {
fmt.Fprintf(os.Stderr, "Error: %v.", e)
os.Exit(0)
}
fmt.Fprintf(os.Stderr, "Writing CPU profile data to %s\n", *cpuprofile)
pprof.StartCPUProfile(profile)
defer pprof.StopCPUProfile()
}
if *inName == "" {
fmt.Fprintln(os.Stderr, "Reading sequences from stdin.")
in = fasta.NewReader(os.Stdin)
} else if in, e = fasta.NewReaderName(*inName); e != nil {
fmt.Fprintf(os.Stderr, "Error: %v.", e)
os.Exit(0)
} else {
fmt.Fprintf(os.Stderr, "Reading sequence from `%s'.\n", *inName)
}
defer in.Close()
if *outName == "" {
fmt.Fprintln(os.Stderr, "Writing output to stdout.")
out = os.Stdout
} else if out, e = os.Create(*outName); e != nil {
fmt.Fprintf(os.Stderr, "Error: %v.", e)
} else {
fmt.Fprintf(os.Stderr, "Writing output to `%s'.\n", *outName)
}
defer out.Close()
totalkmers := make(map[kmerindex.Kmer]float64)
kmerlists := make([]map[kmerindex.Kmer]float64, 0)
seqTable := make([]string, 0)
for {
if sequence, err := in.Read(); err != nil {
break
} else {
var freqs map[kmerindex.Kmer]float64
if kindex, e := kmerindex.New(*k, sequence); e != nil {
fmt.Fprintf(os.Stderr, "Error: %v.\n", e)
fmt.Fprintln(os.Stderr)
os.Exit(0)
} else {
freqs, _ = kindex.NormalisedKmerFrequencies()
kmerlists = append(kmerlists, freqs)
for kmer, freq := range freqs {
totalkmers[kmer] += freq
}
}
seqTable = append(seqTable, string(sequence.ID))
}
}
kmerArray := make([][]float64, 0)
kmerTable := make([]kmerindex.Kmer, 0)
for kmer, _ := range totalkmers {
var count int
for _, kmerlist := range kmerlists {
if kmerlist[kmer] > 0 {
count++
}
}
if count < *lo || float64(count)/float64(len(kmerlists)) > *hi {
continue
}
row := make([]float64, len(kmerlists))
for i, kmerlist := range kmerlists {
row[i] = float64(kmerlist[kmer])
}
kmerArray = append(kmerArray, row)
kmerTable = append(kmerTable, kmer)
}
var kmerMatrix *sparse.Sparse
func() {
defer func() {
if r := recover(); r != nil {
fmt.Fprintf(os.Stderr, "Error: %v.", r)
os.Exit(0)
}
}()
kmerMatrix = sparse.Matrix(kmerArray)
}()
f := func(i, j int, v float64) float64 {
if kmerMatrix.At(i, j) != 0 {
return 1
}
return 0
}
nonZero := kmerMatrix.Apply(f).Sum()
r, c := kmerMatrix.Dims()
density := nonZero / float64(r*c)
if *seed == -1 {
*seed = time.Now().UnixNano()
}
fmt.Fprintf(os.Stderr, "Using %v as random seed.\n", *seed)
rand.Seed(*seed)
rows, cols := kmerMatrix.Dims()
Wo := sparse.Random(rows, *cat, density**sf)
Ho := sparse.Random(*cat, cols, density**sf)
fmt.Fprintf(os.Stderr, "Dimensions of Kmer matrix = (%v, %v)\nDensity = %.3f %%\n%v\n", r, c, (density)*100, kmerMatrix)
W, H, ok := nmf.Factors(kmerMatrix, Wo, Ho, *tol, *iter, time.Duration(*limit)*1e9)
fmt.Fprintf(os.Stderr, "norm(H) = %v norm(W) = %v\n\nFinished = %v\n\n", H.Norm(matrix.Fro), W.Norm(matrix.Fro), ok)
printFeature(out, kmerMatrix, W, H, seqTable, kmerTable, *k)
}
func main() {
var (
in *bufio.Reader
out *bufio.Writer
profile *os.File
e error
)
inName := flag.String("in", "", "Filename for input to be factorised. Defaults to stdin.")
outName := flag.String("out", "", "Filename for output. Defaults to stdout.")
transpose := flag.Bool("t", false, "Transpose columns and rows.")
sep := flag.String("sep", "\t", "Column delimiter.")
cat := flag.Int("cat", 5, "number of categories.")
iter := flag.Int("i", 1000, "iterations.")
rep := flag.Int("rep", 1, "Resample replicates.")
limit := flag.Int("time", 10, "time limit for NMF.")
sf := flag.Float64("sf", 0.01, "factor for sparcity of estimating matrices for NMF.")
tol := flag.Float64("tol", 0.001, "tolerance for NMF.")
threads := flag.Int("threads", 2, "number of threads to use.")
seed := flag.Int64("seed", -1, "seed for random number generator (-1 uses system clock).")
cpuprofile := flag.String("cpuprofile", "", "write cpu profile to this file.")
help := flag.Bool("help", false, "print this usage message.")
flag.Parse()
if *help {
flag.Usage()
os.Exit(1)
}
runtime.GOMAXPROCS(*threads)
sparse.MaxProcs = *threads
fmt.Fprintf(os.Stderr, "Using %d threads.\n", runtime.GOMAXPROCS(0))
if *cpuprofile != "" {
if profile, e = os.Create(*cpuprofile); e != nil {
fmt.Fprintf(os.Stderr, "Error: %v.", e)
os.Exit(0)
}
fmt.Fprintf(os.Stderr, "Writing CPU profile data to %s\n", *cpuprofile)
pprof.StartCPUProfile(profile)
defer pprof.StopCPUProfile()
}
if *inName == "" {
fmt.Fprintln(os.Stderr, "Reading table from stdin.")
in = bufio.NewReader(os.Stdin)
} else if f, err := os.Open(*inName); err != nil {
fmt.Fprintf(os.Stderr, "Error: %v.", err)
os.Exit(0)
} else {
defer f.Close()
in = bufio.NewReader(f)
fmt.Fprintf(os.Stderr, "Reading table from `%s'.\n", *inName)
}
if *outName == "" {
fmt.Fprintln(os.Stderr, "Writing output to stdout.")
out = bufio.NewWriter(os.Stdout)
} else if f, err := os.Create(*outName); err != nil {
fmt.Fprintf(os.Stderr, "Error: %v.", err)
os.Exit(0)
} else {
defer f.Close()
out = bufio.NewWriter(f)
fmt.Fprintf(os.Stderr, "Writing output to `%s'.\n", *outName)
}
defer out.Flush()
var colNames, rowNames []string
array := make([][]float64, 0)
if line, err := in.ReadString('\n'); err != nil {
fmt.Fprintln(os.Stderr, "No table to read\n")
os.Exit(0)
} else {
line = strings.TrimSpace(line)
colNames = strings.Split(line, "\t")
colNames = colNames[1:]
}
for count := 1; ; count++ {
if line, err := in.ReadString('\n'); err != nil {
break
} else {
line = strings.TrimSpace(line)
if row := strings.Split(line, *sep); len(row) != len(colNames)+1 {
fmt.Fprintf(os.Stderr, "Table row mismatch at line %d.\n", count)
os.Exit(0)
} else {
rowData := make([]float64, len(row)-1)
for i, val := range row[1:] {
if rowData[i], e = strconv.ParseFloat(val, 64); e != nil {
fmt.Fprintf(os.Stderr, "Float conversion error %v at line %d element %d.\n", e, count, i)
os.Exit(0)
}
}
rowNames = append(rowNames, row[0])
array = append(array, rowData)
}
}
}
var dataMatrix *sparse.Sparse
func() {
defer func() {
if r := recover(); r != nil {
fmt.Fprintf(os.Stderr, "Error: %v.", r)
os.Exit(0)
}
}()
dataMatrix = sparse.Matrix(array)
}()
f := func(i, j int, v float64) float64 {
if dataMatrix.At(i, j) != 0 {
return 1
}
return 0
}
nonZero := dataMatrix.Apply(f).Sum()
if *transpose {
colNames, rowNames = rowNames, colNames
dataMatrix = dataMatrix.T()
}
r, c := dataMatrix.Dims()
density := nonZero / float64(r*c)
if *seed == -1 {
*seed = time.Now().UnixNano()
}
fmt.Fprintf(os.Stderr, "Using %v as random seed.\n", *seed)
rand.Seed(*seed)
rows, cols := dataMatrix.Dims()
fmt.Fprintf(os.Stderr, "Dimensions of matrix = (%v, %v)\nDensity = %.3f %%\n%v\n", r, c, (density)*100, dataMatrix)
for run := 0; run < *rep; run++ {
if *rep > 1 {
fmt.Fprintf(os.Stderr, "Replicate #%d\n", run+1)
}
Wo := sparse.Random(rows, *cat, density**sf)
Ho := sparse.Random(*cat, cols, density**sf)
W, H, ok := nmf.Factors(dataMatrix, Wo, Ho, *tol, *iter, time.Duration(*limit)*1e9)
fmt.Fprintf(os.Stderr, "norm(H) = %v norm(W) = %v\n\nFinished = %v\n\n", H.Norm(matrix.Fro), W.Norm(matrix.Fro), ok)
printFeature(out, run, dataMatrix, W, H, rowNames, colNames)
}
}