将 dataframe 的所有行与新变量合并为一行

Question

I have a large dataset of segmented cells at different timepoints of a cell culture. My end goal is to use a classifier to be able to separate cells from the early time points from later cells by using their shape. However, what I need to use for this is the overall composition of the shape of the cells (ie a later timepoint will contain more larger cells that are less round).

My data looks like this, passage being the "class" variable (1:8).

     Area Circ.       X      Y  Major  Minor   Angle    AR Round Solidity Passage
1 270.606 0.476 369.677 11.832 21.497 16.028 177.550 1.341 0.746    0.902       1
2 260.733 0.652 219.469 14.233 18.847 17.614  51.695 1.070 0.935    0.948       1
3 444.248 0.682  70.619 24.071 27.845 20.313  54.227 1.371 0.730    0.953       1
4 236.565 0.607 409.612 21.472 18.800 16.022 110.348 1.173 0.852    0.939       1
5 291.237 0.376 547.529 19.330 30.212 12.274 178.844 2.462 0.406    0.915       1
6 201.690 0.662 202.990 20.799 16.457 15.604  54.949 1.055 0.948    0.949       1

My approach now is to sample n amount of cells from one passage and train the classifier with this.

#number of cells in image
num_draw = 15
all_samples <- list()
for (i in 1:8){
  sample_list <- list()
  for (j in 1:1000){
    samples = filter(df, df$Passage == i)
    sample_list[[j]] = samples[sample(nrow(samples), 20), ]
  }
  all_samples[[i]] <- sample_list
}

However, most classifiers seem to only take data that has one observation per row, so I think my best shot is to combine all cells in a sample to a single row, keeping all the variables/columns (probably as Area1, Area2, Area3,...) except passage, which is combined to one column, then combining all observations back together into one dataframe.

Is there a quick way of doing this? Or can you recommend me a classifier that can take n instances of one class as one observation?

Answer 1

I'm not sure why you would want to do this because I think it will make it much harder to work with your data. Maybe think about pivot_longer instead, and using group_by and summarize. Anyways here is my solution. Best of luck!

library(tidyr)
library(dplyr)

#create data
df <- data.frame(Area = rnorm(4), circ = rnorm(4), x = rnorm(4), passage = c("1", "1", "2", "2"))

#spread data wide based off number of observations of passage
df <- group_by(df, passage) %>%
  mutate(reference = row_number()) %>% #creates reference for each obs of passage
  pivot_wider(id_cols = passage, names_from = reference, values_from = Area:x)