selOutliers.R 7.48 KB
Newer Older
dmattek's avatar
dmattek committed
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
#
# Time Course Inspector: Shiny app for plotting time series data
# Author: Maciej Dobrzynski
#
# This is the module of a Shiny web application.
# Outlier identification, selection

# UI-remove-outliers ----
modSelOutliersUI = function(id, label = "Outlier Selection") {
  ns <- NS(id)
  
  tagList(
    h4(
      "Remove outliers"
    ),
    fluidRow(
      column(2, 
             numericInput(ns('numOutliersPerc'),
                         label = '% of data',
                         min = 0,
                         max = 100,
                         value = 0, 
23
24
                         step = 0.05, width = '100px'),
             checkboxInput(ns('chBtrajInter'), 'Interpolate gaps?', value = F)
dmattek's avatar
dmattek committed
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
      ),
      column(2, 
             radioButtons(ns('rbOutliersType'), 
                          label = 'From', 
                          choices = c('top' = 'top', 'middle' = 'mid', 'bottom' = 'bot'))
             ),
      column(3,
             sliderInput(ns('slOutliersGapLen'),
                         label = 'Remove tracks with gaps equal to or longer than',
                         min = 1,
                         max = 10,
                         value = 1, 
                         step = 1)
      ),
      column(3,
             downloadButton(ns('downOutlierCSV'), label = 'CSV with outlier IDs'),
             htmlOutput(ns("txtOutliersPerc"))
      )
    )
  )
}

# Server-remove-outliers ----
modSelOutliers = function(input, output, session, in.data) {

  # reactive counter to hold number of tracks before and after outlier removal
  nCellsCounter <- reactiveValues(
    nCellsOrig = 0,
    nCellsAfter = 0,
    nOutlierTpts = 0
  )
  
  # reactive vector with cell ids
  vOut = reactiveValues(
    id = NULL
  )

62
63
  
  
dmattek's avatar
dmattek committed
64
65
  # Display number of tracks and outliers  
  output$txtOutliersPerc <- renderText({ 
66
    cat(file = stdout(), 'modSelOutliers: txtOutliersPerc\n')
dmattek's avatar
dmattek committed
67
    
68
      sprintf('<b>%d total track(s)<br>%d outlier track(s)<br>%d outlier point(s)</b><br>', 
dmattek's avatar
dmattek committed
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
            nCellsCounter[['nCellsOrig']], 
            nCellsCounter[['nCellsOrig']] - nCellsCounter[['nCellsAfter']],
            nCellsCounter[['nOutlierTpts']])
    })
  
  # button for downloading CSV with ids of removed tracks
  output$downOutlierCSV <- downloadHandler(
    filename = FCSVOUTLIERS,
    content = function(file) {
      loc.dt = vOut[['id']]
      
      if (is.null(loc.dt))
        return(NULL)
      else
        write.csv(unique(loc.dt[, (COLID), with = F]), file, row.names = FALSE, quote = F)
    }
  )
  
# Identify outliers and remove them from dt
  dtReturn = reactive({ 
89
    cat(file = stdout(), 'modSelOutliers: dtReturn\n')
dmattek's avatar
dmattek committed
90
91
92
93
94
95
96
    
    loc.out = in.data()
    
    if (is.null(loc.out)) {
      return(NULL)
    }

97
98
99
    # store the number of trajectories before prunning
    nCellsCounter[['nCellsOrig']] = length(unique(loc.out[['id']]))
    
100
101
    # Remove outliers if the field with percentage of data to remove is greater than 0
    if (input$numOutliersPerc > 0) {
dmattek's avatar
dmattek committed
102
      
103
      # scale all measurement points      
dmattek's avatar
dmattek committed
104
105
106
107
108
      loc.out[, y.sc := scale(get(COLY))]  

      # Identify outlier points
      # In the UI, user selectes percentage of data to remove from the bottom, middle, or top part.
      # loc.outpts stores outlier points
109
      # warning: quantile type = 3: SAS definition: nearest even order statistic.
dmattek's avatar
dmattek committed
110
      switch(input$rbOutliersType,
111
112
113
114
        'top' = {loc.outpts = loc.out[ y.sc > quantile(y.sc, 1 - input$numOutliersPerc * 0.01, na.rm = T, type = 3)]},
        'mid' = {loc.outpts = loc.out[ y.sc < quantile(y.sc, input$numOutliersPerc * 0.005, na.rm = T, type = 3) | 
                                     y.sc > quantile(y.sc, 1 - input$numOutliersPerc * 0.005, na.rm = T, type = 3)]},
        'bot' = {loc.outpts = loc.out[ y.sc < quantile(y.sc, input$numOutliersPerc * 0.01, na.rm = T, type = 3)]}
dmattek's avatar
dmattek committed
115
116
      )
      
117
118
119
120
121
      if (DEB) {
        cat(file = stdout(), 'selOutliers.dtReturn: Outlier points:\n')
        print(loc.outpts)
      }
        
dmattek's avatar
dmattek committed
122
123
124
125
126
127
      if (input$slOutliersGapLen > 1) {
        # remove tracks with gaps longer than the value set in slOutliersGapLen
        # shorter gaps are interpolated linearly
        
        # add index column per trajecory
        loc.out[, (COLIDX) := 1:.N, by = c(COLID)]
128

dmattek's avatar
dmattek committed
129
130
131
        # remove single outlier points (anti-join)
        # From: https://stackoverflow.com/a/46333620/1898713
        loc.out = loc.out[!loc.outpts, on = names(loc.outpts)]
132

dmattek's avatar
dmattek committed
133
134
135
136
137
138
139
        # calculate diff on index column to see the length of gaps due to removed points
        # the value of that column corresponds to the gap length (hence the "-1")
        loc.out[, (COLIDXDIFF) := c(1, diff(get(COLIDX))) - 1, by = c(COLID)]

        # get track ids where the max gap is equal to or longer than the threshold
        loc.idgaps = loc.out[, max(get(COLIDXDIFF)), by = c(COLID)][V1 >= input$slOutliersGapLen, get(COLID)]
        
140
141
142
143
144
145
        if (DEB) {
          cat(file = stdout(), '\nselOutliers.dtReturn: Track IDs with max gap >= threshold:\n')
          if (length(loc.idgaps) > 0)
            print(loc.idgaps) else
              cat("none\n")
        }
dmattek's avatar
dmattek committed
146
        
147
148
149
150
        # remove outlier tracks with gaps longer than the value set in slOutliersGapLen
        if (length(loc.idgaps) > 0)
          loc.out = loc.out[!(get(COLID) %in% unique(loc.idgaps))]

dmattek's avatar
dmattek committed
151
152
        # clean
        loc.out[, c(COLIDX, COLIDXDIFF) := NULL]
153
154
155
156
157
158
159
160
161

        # interpolate gaps due to outliers
        if (input$chBtrajInter) {
          # fill removed outliers with NA's
          setkeyv(loc.out, c(COLGR, COLID, COLRT))
          loc.out = loc.out[setkeyv(loc.out[, .(seq(min(get(COLRT), na.rm = T), max(get(COLRT), na.rm = T), 1)), by = c(COLGR, COLID)], c(COLGR, COLID, 'V1'))]

          # x-check: print all rows with NA's
          if (DEB) {
162
            cat(file = stdout(), '\nselOutliers.dtReturn: Rows with NAs to interpolate:\n')
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
            print(loc.out[rowSums(is.na(loc.out)) > 0, ])
          }
          
          # NA's may be already present in the dataset'.
          # Interpolate (linear) them with na.interpolate as well
          if( (COLPOSX %in% names(loc.out)) & (COLPOSY %in% names(loc.out)) )
            s.cols = c(COLY, COLPOSX, COLPOSY)
          else
            s.cols = c(COLY)
          
          
          # Apparently the loop is faster than lapply+SDcols
          for(col in s.cols) {
            # Interpolated columns should be of type numeric (float)
            # This is to ensure that interpolated columns are of porper type.
            data.table::set(loc.out, j = col, value = as.numeric(loc.out[[col]]))
            
            loc.out[, (col) := na.interpolation(get(col)), by = c(COLID)]        
          }
        } 
dmattek's avatar
dmattek committed
183
184
      } else {
        # remove outlier tracks with gaps of length 1 time point
185
        # !(input$slOutliersGapLen > 1)
dmattek's avatar
dmattek committed
186
187
188
189
190
        loc.out = loc.out[!(get(COLID) %in% unique(loc.outpts[[COLID]]))]
      }

      # clean
      loc.out[, y.sc := NULL]
191

dmattek's avatar
dmattek committed
192
193
194
      
      # store a vector of outlier timepoints with the corresponding IDs
      vOut[['id']] = loc.outpts
195
196
197
198
199
    } else {
      # no outlier removal
      # !(input$numOutliersPerc > 0)
      loc.outpts = NULL
      vOut = NULL
dmattek's avatar
dmattek committed
200
    }
201
202
203
204
205
206
207

    # count number of trajectories after removing outlier tracks
    nCellsCounter[['nCellsAfter']] = length(unique(loc.out[[COLID]]))
    
    # count number of outlier points
    nCellsCounter[['nOutlierTpts']] = length(loc.outpts[[COLID]])
    cat(sprintf("%d outlier tpts\n", nCellsCounter[['nOutlierTpts']]))
dmattek's avatar
dmattek committed
208
209
    
    # return cleaned dt
210
211
212
    if (nrow(loc.out) < 1)
      return(NULL) else
        return(loc.out)
dmattek's avatar
dmattek committed
213
214
215
216
217
    
  })
  
  return(dtReturn)
}