Summarizing subsets of data#

Assume we want to summarize our data, e.g. by splitting it into groups according to filename and computing mean intensity measurements for these groups. This will give us a smaller table with summarized measurements per file.

See also

import pandas as pd
import numpy as np

To demonstate the example, we load a table which contains shape measurements of many objects that have been segmented in multiple files of the Broad Bioimage Benchmark Collection BBB0007 dataset from Jones et al., Proc. ICCV Workshop on Computer Vision for Biomedical Image Applications, 2005).

df = pd.read_csv('../../data/BBBC007_analysis.csv')
df
area intensity_mean major_axis_length minor_axis_length aspect_ratio file_name
0 139 96.546763 17.504104 10.292770 1.700621 20P1_POS0010_D_1UL
1 360 86.613889 35.746808 14.983124 2.385805 20P1_POS0010_D_1UL
2 43 91.488372 12.967884 4.351573 2.980045 20P1_POS0010_D_1UL
3 140 73.742857 18.940508 10.314404 1.836316 20P1_POS0010_D_1UL
4 144 89.375000 13.639308 13.458532 1.013432 20P1_POS0010_D_1UL
... ... ... ... ... ... ...
106 305 88.252459 20.226532 19.244210 1.051045 20P1_POS0007_D_1UL
107 593 89.905565 36.508370 21.365394 1.708762 20P1_POS0007_D_1UL
108 289 106.851211 20.427809 18.221452 1.121086 20P1_POS0007_D_1UL
109 277 100.664260 20.307965 17.432920 1.164920 20P1_POS0007_D_1UL
110 46 70.869565 11.648895 5.298003 2.198733 20P1_POS0007_D_1UL

111 rows × 6 columns

Grouping by filename#

We will now group the table by image filename.

grouped_df = df.groupby('file_name')
grouped_df

<pandas.core.groupby.generic.DataFrameGroupBy object at 0x000002DC95CF2520>

From this grouped_df object we can derive basic statistics, for example the mean of all numeric columns.

summary_df = grouped_df.mean(numeric_only = True)
summary_df
area intensity_mean major_axis_length minor_axis_length aspect_ratio
file_name
20P1_POS0007_D_1UL 300.859375 95.889956 22.015742 17.132505 1.316197
20P1_POS0010_D_1UL 253.361702 96.745373 20.120268 15.330923 1.402934

The outputted data frame has the mean values of all quantities, including the intensities that we wanted. Note that this data frame has ‘filename’ as the name of the row index. To convert it back to a normal table with a numeric index columm, we can use the reset_index() method.

summary_df.reset_index()
file_name area intensity_mean major_axis_length minor_axis_length aspect_ratio
0 20P1_POS0007_D_1UL 300.859375 95.889956 22.015742 17.132505 1.316197
1 20P1_POS0010_D_1UL 253.361702 96.745373 20.120268 15.330923 1.402934

Note, though, that this was not done in-place. summary_df still has an index labeled round. If you want to update your table, you have to explicitly do so with an assignment operator.

summary_df = summary_df.reset_index()
summary_df
file_name area intensity_mean major_axis_length minor_axis_length aspect_ratio
0 20P1_POS0007_D_1UL 300.859375 95.889956 22.015742 17.132505 1.316197
1 20P1_POS0010_D_1UL 253.361702 96.745373 20.120268 15.330923 1.402934