selOutliers.R 11.6 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
                         step = 0.05, width = '100px'),
24
             checkboxInput(ns('chBtrajInter'), 'Interpolate gaps', value = F)
dmattek's avatar
dmattek committed
25
26
27
28
      ),
      column(2, 
             radioButtons(ns('rbOutliersType'), 
                          label = 'From', 
Maciej Dobrzynski's avatar
Maciej Dobrzynski committed
29
                          choices = c('top' = 'top', 'top & bottom' = 'mid', 'bottom' = 'bot'))
dmattek's avatar
dmattek committed
30
31
32
33
34
35
36
37
38
39
40
41
42
             ),
      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"))
      )
43
44
45
    ),
    checkboxInput(ns('chBplotDist'), 'Plot data distribution', value = F),
    uiOutput(ns('uiDistPlot'))
dmattek's avatar
dmattek committed
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
  )
}

# 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
  )

64
65
  
  
dmattek's avatar
dmattek committed
66
67
  # Display number of tracks and outliers  
  output$txtOutliersPerc <- renderText({ 
68
    cat(file = stdout(), 'modSelOutliers: txtOutliersPerc\n')
dmattek's avatar
dmattek committed
69
    
70
      sprintf('<b>%d total track(s)<br>%d outlier track(s)<br>%d outlier point(s)</b><br>', 
dmattek's avatar
dmattek committed
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)
    }
  )
  
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
  # Plot of value distribution
  output$uiDistPlot <- renderUI({
    ns <- session$ns
    
    if (input$chBplotDist) {

      locDT = in.data()
      
      if (is.null(locDT)) {
        return(NULL)
      }

      output$densPlot = renderPlot({

        # main density plot
        locP = ggplot(locDT, aes_string(x = COLY)) +
          geom_density()
        
        # Shade regions of the density plot according to
        # value set in input$numOutliersPerc.
        
        # extract data from density plot
        locDTtmp = as.data.table(ggplot_build(locP)$data[[1]])
        
        # shade region on the right
        if (input$rbOutliersType == 'top') {
          
          # find position of the right boundary
          locQuantR = quantile(locDT[[COLY]], 
                               1 - input$numOutliersPerc * 0.01, 
                               na.rm = T, 
                               type = 3)
          
          # select only those points of the density plot right to the right boundary
          locDTtmpSub = locDTtmp[x > locQuantR]
          
          # add shaded RIGHT region to the plot
          if (nrow(locDTtmpSub) > 0 )
            locP = locP + 
            geom_area(data = locDTtmpSub, aes(x=x, y=y), fill="red") +
            geom_vline(xintercept = locQuantR, linetype = 'dashed', color = 'red')
        } else 
          # shade region on the left
          if (input$rbOutliersType == 'bot') {
            
            # find position of the right boundary
            locQuantL = quantile(locDT[[COLY]], 
                                 input$numOutliersPerc * 0.01, 
                                 na.rm = T, 
                                 type = 3)
            
            # select only those points of the density plot left to the left boundary
            locDTtmpSub = locDTtmp[x < locQuantL]
            
            # add shaded LEFT region to the plot
            if (nrow(locDTtmpSub) > 0 )
              locP = locP + 
              geom_area(data = locDTtmpSub, aes(x=x, y=y), fill="red") +
              geom_vline(xintercept = locQuantL, linetype = 'dashed', color = 'red')
            
          } else 
            # shade region on the left
            if (input$rbOutliersType == 'mid') {
              
              # find position of the right boundary
              locQuantR = quantile(locDT[[COLY]], 
                                   1 - input$numOutliersPerc * 0.005, 
                                   na.rm = T, 
                                   type = 3)
              
              # find position of the left boundary
              locQuantL = quantile(locDT[[COLY]], 
                                   input$numOutliersPerc * 0.005, 
                                   na.rm = T, 
                                   type = 3)
              
              # select only those points of the density plot left or right of the boundaries
              locDTtmpSubL = locDTtmp[x < locQuantL]
              locDTtmpSubR = locDTtmp[x > locQuantR]
              
              # add shaded LEFT region to the plot
              if (nrow(locDTtmpSubL) > 0 )
                locP = locP + 
                geom_area(data = locDTtmpSubL, aes(x=x, y=y), fill="red") +
                geom_vline(xintercept = locQuantL, linetype = 'dashed', color = 'red')
              
              
              if (nrow(locDTtmpSubR) > 0 )
                locP = locP + 
                geom_area(data = locDTtmpSubR, aes(x=x, y=y), fill="red") +
                geom_vline(xintercept = locQuantR, linetype = 'dashed', color = 'red')
            }
        
        locP = locP +
          xlab('Measurement value') +
          LOCggplotTheme(in.font.base = PLOTFONTBASE, 
                         in.font.axis.text = PLOTFONTAXISTEXT, 
                         in.font.axis.title = PLOTFONTAXISTITLE, 
                         in.font.strip = PLOTFONTFACETSTRIP, 
                         in.font.legend = PLOTFONTLEGEND)
        
        return(locP)
        
      })
      
    } else
      return(NULL)
    
    plotOutput(ns('densPlot'))
  })
  
dmattek's avatar
dmattek committed
200
201
# Identify outliers and remove them from dt
  dtReturn = reactive({ 
202
    cat(file = stdout(), 'modSelOutliers: dtReturn\n')
dmattek's avatar
dmattek committed
203
204
205
206
207
208
209
    
    loc.out = in.data()
    
    if (is.null(loc.out)) {
      return(NULL)
    }

210
211
212
    # store the number of trajectories before prunning
    nCellsCounter[['nCellsOrig']] = length(unique(loc.out[['id']]))
    
213
214
    # Remove outliers if the field with percentage of data to remove is greater than 0
    if (input$numOutliersPerc > 0) {
dmattek's avatar
dmattek committed
215
      
216
      # scale all measurement points      
dmattek's avatar
dmattek committed
217
218
219
220
221
      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
222
      # warning: quantile type = 3: SAS definition: nearest even order statistic.
dmattek's avatar
dmattek committed
223
      switch(input$rbOutliersType,
224
225
226
227
        'top' = {loc.outpts = loc.out[ y.sc > quantile(y.sc, 1 - input$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
228
229
      )
      
230
231
232
233
234
      if (DEB) {
        cat(file = stdout(), 'selOutliers.dtReturn: Outlier points:\n')
        print(loc.outpts)
      }
        
dmattek's avatar
dmattek committed
235
236
237
238
239
240
      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)]
241

dmattek's avatar
dmattek committed
242
243
244
        # remove single outlier points (anti-join)
        # From: https://stackoverflow.com/a/46333620/1898713
        loc.out = loc.out[!loc.outpts, on = names(loc.outpts)]
245

dmattek's avatar
dmattek committed
246
247
248
249
250
251
252
        # 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)]
        
253
254
255
256
257
258
        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
259
        
260
261
262
263
        # 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
264
265
        # clean
        loc.out[, c(COLIDX, COLIDXDIFF) := NULL]
266
267
268
269
270
271
272
273
274

        # 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) {
275
            cat(file = stdout(), '\nselOutliers.dtReturn: Rows with NAs to interpolate:\n')
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
            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
296
297
      } else {
        # remove outlier tracks with gaps of length 1 time point
298
        # !(input$slOutliersGapLen > 1)
dmattek's avatar
dmattek committed
299
300
301
302
303
        loc.out = loc.out[!(get(COLID) %in% unique(loc.outpts[[COLID]]))]
      }

      # clean
      loc.out[, y.sc := NULL]
304

dmattek's avatar
dmattek committed
305
306
307
      
      # store a vector of outlier timepoints with the corresponding IDs
      vOut[['id']] = loc.outpts
308
309
310
311
312
    } else {
      # no outlier removal
      # !(input$numOutliersPerc > 0)
      loc.outpts = NULL
      vOut = NULL
dmattek's avatar
dmattek committed
313
    }
314
315
316
317
318
319
320

    # 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
321
322
    
    # return cleaned dt
323
324
325
    if (nrow(loc.out) < 1)
      return(NULL) else
        return(loc.out)
dmattek's avatar
dmattek committed
326
327
328
329
330
    
  })
  
  return(dtReturn)
}