// Function turns file into slice with feature symbols
// - feature == 0 -> position in paragraph
// - feature == 1 -> sentence length // now for else
func CreateObservationSequence(filename string, states int) [][]int {
output := make([][]int, 0, 0)
s_number := 0
reader_full := GetReader(filename)
for bpar, e := reader_full.ReadBytes('\n'); e == nil; bpar, e = reader_full.ReadBytes('\n') {
var sentences []string
par := nlptk.Paragraph{0, string(bpar[:len(bpar)-1])}
if sentences = par.GetParts(); len(par.Text) <= 1 || len(sentences) == 0 {
continue
}
for i, s := range sentences {
sentence := nlptk.Sentence{s_number, s[:len(s)-1]}
output = append(output, []int{i + 1, len(sentence.GetParts())}) // summary state
}
if s_number++; 2*s_number >= states {
break
}
}
return output
}
// Extracts, trims from special signs and counts "bare" words in learning set.
// Result dictionary is sent to channel.
func WordCount(filename string, dict chan map[string]int) {
word_counter := make(map[string]int)
file, err := os.Open(SETDIR + filename)
if err != nil {
fmt.Println("Error reading file", filename)
os.Exit(1)
}
reader := bufio.NewReader(file)
for bpar, e := reader.ReadBytes('\n'); e == nil; bpar, e = reader.ReadBytes('\n') {
paragraph := nlptk.Paragraph{0, string(bpar)}
sentences := paragraph.GetParts()
for _, sentence := range sentences {
s := nlptk.Sentence{0, sentence}
words := s.GetParts()
if len(words) == 0 {
continue
}
for _, word := range words {
word_counter[word]++
word_counter["TOTAL"]++
}
}
}
dict <- word_counter
}
// Analyze full text and summarization to prepare observations:
// - vectors of features
// - binary table of sentence in summarization presence
func ObserveFile(filename, full_dir, summ_dir string, states int) [][]int {
reader_full := GetReader(full_dir + filename)
reader_summ := GetReader(summ_dir + filename)
s_counter := make([][]int, 0, 0)
s_number, p_number := 0, 0
spar, _ := reader_summ.ReadBytes('\n')
summarization := nlptk.Paragraph{p_number, string(spar)}
sum_sentences := summarization.Text
for bpar, e := reader_full.ReadBytes('\n'); e == nil; bpar, e = reader_full.ReadBytes('\n') {
var sentences []string
paragraph := nlptk.Paragraph{p_number, string(bpar[:len(bpar)-1])}
if sentences = paragraph.GetParts(); len(paragraph.Text) <= 1 || len(sentences) == 0 {
continue
}
for i, s := range sentences {
sentence := nlptk.Sentence{s_number, s[:len(s)-1]}
if strings.Contains(sum_sentences, s[:len(s)-1]) {
// summary
s_counter = append(s_counter, []int{0, 0})
s_counter = append(s_counter, []int{i + 1, len(sentence.GetParts())})
} else {
// non-summary
s_counter = append(s_counter, []int{i + 1, len(sentence.GetParts())})
s_counter = append(s_counter, []int{0, 0})
}
if s_number++; 2*s_number >= states {
return s_counter
}
}
p_number++
}
fmt.Println("Sequence", filename, s_counter)
return s_counter
}
// Prints sequence of states (appear, not appear) given by slice
func PrintSequence(filename string, sequence []int) string {
output := make([]string, 0, 0)
s_number := 0
reader_full := GetReader(filename)
writer := GetWriter(filename)
for bpar, e := reader_full.ReadBytes('\n'); e == nil; bpar, e = reader_full.ReadBytes('\n') {
paragraph := nlptk.Paragraph{0, string(bpar[:len(bpar)-1])}
var sentences []string
// check if paragraph is empty
if len(paragraph.Text) <= 1 {
continue
}
// just in case avoid kinky sentences
if sentences = paragraph.GetParts(); len(sentences) == 0 {
continue
}
for _, s := range sentences {
for _, v := range sequence {
if v == 2*s_number+1 {
output = append(output, s)
writer.Write([]byte(s))
}
}
if s_number++; 2*s_number+1 >= len(sequence) {
writer.Flush()
return strings.Join(output, ". ")
}
}
}
writer.Flush()
return strings.Join(output, ". ")
}