// Add a slice of stories to the stories FIFO
func addStories(s []*story.Story) {
stories.mutex.Lock()
defer stories.mutex.Unlock()
for _, story := range s {
stories.add(story)
}
config.Debug("Added ", len(s), "new stories")
config.Debug("start =", stories.start, ", end =", stories.end)
}
// Indicate that a user has browsed up to the given storyid
func markBrowsed(session *Session, storyid int64) {
// We are about to access stories
stories.mutex.RLock()
defer stories.mutex.RUnlock()
if storyid < stories.start || storyid < session.haveBrowsed {
config.Debug("Not marking browsed ", storyid, ", ",
stories.start, ", ", session.haveBrowsed)
session.filtered = session.filtered[0:0]
session.unfiltered = session.unfiltered[0:0]
return
}
if storyid > stories.end {
log.Println("User has browsed past the last story", storyid, ">", stories.end)
session.filtered = session.filtered[0:0]
session.unfiltered = session.unfiltered[0:0]
return
}
// Loop through the filtered and unfiltered stories and mark
// the unread ones as uninteresting
for _, s := range session.filtered {
i := s.Id
if i < session.haveBrowsed {
config.Debug("Not uninteresting: ", i)
continue
}
if !session.haveRead[i] && !session.haveIgnored[i] {
session.classifyStory(i, Uninteresting)
session.haveIgnored[i] = true
}
}
for _, s := range session.unfiltered {
i := s.Id
if i < session.haveBrowsed {
config.Debug("Not uninteresting: ", i)
continue
}
if !session.haveRead[i] && !session.haveIgnored[i] {
session.classifyStory(i, Uninteresting)
}
}
// Clear the filtered and unfiltered stories
session.filtered = session.filtered[0:0]
session.unfiltered = session.unfiltered[0:0]
// Keep track of how far the user has browsed
session.haveBrowsed = storyid - 1
}
// Add a story to the fifo
func (f *fifo) add(s *story.Story) {
// The fifo relies on the assumption that storyids always
// increase
if s.Id < f.end {
log.Panicln("Non-increasing story id added to fifo: ",
s.Id, ", ", f.start, ", ", f.end)
}
if len(f.index) == 0 {
f.start = s.Id
f.index = append(f.index, s)
f.end = s.Id
} else if len(f.index) == cap(f.index) {
f.index[f.head] = s
f.head = (f.head + 1) % cap(f.index)
f.start = f.index[f.head].Id
f.end = s.Id
} else {
f.index = append(f.index, s)
f.end = s.Id
}
config.Debug("Added story: ", s.Id, ", ", f.start, "-", f.end)
}
// Classify the given word list, returns the id of the class
func (c *Classifier) Classify(words []string) int {
// Prefilter the words
filtered := c.Prefilter(words)
weights := make([]float64, len(c.Classes))
// Loop through the classes
for i := range c.Classes {
config.Debug("Weighting class ", i)
weights[i] = c.weight(c.Classes[i], filtered)
}
// Find the heaviest class
h := 0
hw := weights[h]
for j, w := range weights {
if w > hw {
hw = w
h = j
}
}
return h
}
// Get the session from the cookie in the request
// Creates a new session and cookie if none can be found
func getSession(w http.ResponseWriter, req *http.Request) (*Session, bool) {
cookie, err := req.Cookie("id")
if err == nil {
s, ok := sessionSync(cookie.Value)
if ok {
return s, ok
}
log.Println("Invalid session cookie presented: ", cookie.Value)
}
// Create a session
s := newSession()
config.Debug("Creating new session ", s.id)
sessions.Create(s)
// Write the session to the HTTP response
newcookie := http.Cookie{
Name: "id",
Value: s.id,
Path: "/",
MaxAge: math.MaxInt32,
HttpOnly: true}
w.Header().Add("Set-Cookie", newcookie.String())
s.mutex.Lock()
return s, true
}
// Get interesting stories starting at the given story
func interesting(session *Session, start int64) []*story.Story {
// We are about to access stories
stories.mutex.RLock()
defer stories.mutex.RUnlock()
ret := make([]*story.Story, 0, interestingPerPage)
if start > stories.end {
return ret
}
for i := start; i < stories.end; i++ {
if session.haveRead[i] || session.haveIgnored[i] {
continue
}
story, ok := stories.get(i)
if !ok {
break
}
class := session.classifier.Classify(story.Wordlist)
if class == Interesting {
config.Debug("Interesting:", story.Rss.Title)
ret = append(ret, story)
if len(ret) == interestingPerPage {
break
}
}
}
return ret
}
// Calculate the weight of the given wordlist for the given class
func (c *Classifier) weight(class Class, words []string) float64 {
// Handle pathological cases
if c.Total == 0 {
return 0
} else if class.Count == 0 {
return -500
}
//log.Println("Vocabulary:", class.vocabulary)
// Calculate the prior, log(P(class))
//prior := math.Log2(float64(class.Count) / float64(c.Total))
//config.Debug("prior ", prior)
prior := class.Prior
// TODO: improve the smoothing - currently wholly unknown titles get
// classified as interesting
// Calculate the log likelihood log(P(words|class))
ll := prior
k := float64(c.Words)
for _, w := range words {
occurs := class.Vocabulary[w]
// Calculate the probability of the word appearing in this class
// but add smoothing to avoid overfitting
// P(w|class) = N(w,class) + 1
// --------------
// N(class) + k
// where:
// k = number of words in the training set
pw := (float64(occurs) + 1) / (float64(class.Count) + k)
config.Debug("P(", w, "|class) = ", pw)
ll += math.Log2(pw)
}
config.Debug("ll =", ll)
return ll
}
// Save a session to the db
func saveSession(c cache.Entry) {
// Convert from the cache format to a session
session, ok := c.(*Session)
if !ok {
log.Fatal("Cannot convert cache.Entry to *Session")
}
config.Debug("Saving session ", session.id)
config.Debug("Writing classifier:")
config.Debug(session.classifier)
config.Debug("Writing classified index: ", session.haveClassified)
config.Debug("Writing browsed index: ", session.haveBrowsed)
// Serialize the session
cbytes, err := session.classifier.Serialise()
if err != nil {
return
}
ibytes, err := serialiseStoryMap(session.haveIgnored)
if err != nil {
return
}
// Write the session
dbs := db.Session{
Id: session.id,
Classifier: cbytes,
HaveIgnored: ibytes,
HaveClassified: session.haveClassified,
HaveBrowsed: session.haveBrowsed}
if session.isNew {
db.CreateSession(&dbs)
} else {
db.WriteSession(&dbs)
}
}
func readFeed(filename string) {
config.Debug("Reading feed: ", filename)
stories, err := rss.Decode(filename)
if err != nil {
log.Println("Cannot decode ", filename, " :", err)
return
}
todo := make([]*story.Story, 0, 64)
for _, s := range stories {
todo = addStory(todo, s)
}
// Add any new stories
session.AddStories(todo)
}
// Get the interesting stories starting at the given story
func FilteredStories(w http.ResponseWriter, req *http.Request, storyid int64) *StoryIndex {
// Get the users session if possible
session, session_ok := getSession(w, req)
if session_ok {
defer session.release()
}
// We are about to access stories
stories.mutex.RLock()
defer stories.mutex.RUnlock()
// Try and start the index at the latest browsed story if no story is specified
start := storyid
if session_ok && storyid == 0 {
if session.haveBrowsed > 0 {
start = session.haveBrowsed + 1
} else {
start = stories.end - DefaultStart
}
config.Debug("Building story index starting from", start)
}
start = max(start, stories.start)
ret := NewStoryIndex()
// Get stories
if !session_ok {
ret.Unfiltered = unfiltered(nil, false, start, storiesPerPage, nil)
} else if len(session.filtered) > 0 || len(session.unfiltered) > 0 {
ret.Filtered = session.filtered
ret.Unfiltered = session.unfiltered
} else {
ret.Filtered = interesting(session, start)
todo := storiesPerPage - len(ret.Filtered)
ret.Unfiltered = unfiltered(session, true, start, todo, storyIdMap(ret.Filtered))
session.filtered = ret.Filtered
session.unfiltered = ret.Unfiltered
}
previousNext(ret, start)
return ret
}
// Read a session from the db
func readSession(key string) (cache.Entry, bool) {
dbs, ok := db.GetSession(key)
if !ok {
log.Println("Failed to read session ", key, " from db")
return nil, false
}
// Convert the session from db format
// Deserialize the classsifier
classifier, err := nbc.Deserialise(dbs.Classifier)
if err != nil {
return nil, false
}
// Get the read stories
read := getReadMap(key)
// Deserialise ignored stories
ignored, err := deserialiseStoryMap(dbs.HaveIgnored)
if err != nil {
return nil, false
}
config.Debug("Deserialised classifier: ", classifier)
// Return the deserialized session
ret := &Session{
id: dbs.Id,
classifier: classifier,
haveRead: read,
haveIgnored: ignored,
haveClassified: dbs.HaveClassified,
haveBrowsed: dbs.HaveBrowsed}
ret.filtered = make([]*story.Story, 0, storiesPerPage)
ret.unfiltered = make([]*story.Story, 0, storiesPerPage)
return ret, true
}
