The first step replicates work from the previous recipe by converting two columns to numeric and eliminating the binary columns. We find the maximum value of each column in step 2. Care needs to be taken here as pandas silently drops columns that it cannot produce a maximum. If this happens, then step 3 will still complete but produce all False values for each column without an available maximum.

Step 4 uses the any method to scan across each row in search of at least one True value. Any row with at least one True value contains a maximum value for a column. We sum up the resulting boolean Series in step 5 to determine how many rows contain a maximum. Somewhat unexpectedly, there are far more rows than columns. Step 6 gives insight on why this happens. We take a cumulative sum of the output from step 3 and detect the total number of rows that equal the maximum for each column.

Many colleges have 100% of their student population as only a single race. This is by far the largest contributor to the multiple rows with maximums. As you can see, there is only one row with a maximum value for both SAT score columns and undergraduate population, but several of the race columns have a tie for the maximum.

Our goal is to find the first row with the maximum value. We need to take the cumulative sum once more so that each column has only a single row equal to 1. Step 8 formats the code to have one method per line and runs the any method exactly as it was done in step 4. If this step is successful, then we should have no more True values than the number of columns. Step 9 asserts that this is true.

To validate that we have found the same columns as idxmax in the previous columns, we use boolean selection on has_row_max2 with itself. The columns will be in a different order so we convert the sequence of column names to sets, which are inherently unordered to compare equality.