Aggregations & axes

Often you don't want every number — you want a summary: the total, the average, the biggest. NumPy has these built in, and one small argument, axis, controls whether you summarize the whole array or just its rows or columns.

Reducing the whole array

Call these on any array to collapse it to a single number:

v = np.array([10, 20, 30, 40, 50])

v.sum()    # 150
v.mean()   # 30.0
v.min()    # 10
v.max()    # 50
v.std()    # 14.142135...   — standard deviation

The axis argument

On a 2-D array, axis picks the direction to collapse. Think of it as "which dimension disappears": axis=0 collapses the rows (leaving one value per column), axis=1 collapses the columns (one value per row):

grades = np.array([[80, 90, 70],     # student 0
                 [60, 75, 95],     # student 1
                 [100, 85, 90]])   # student 2

grades.sum()             # 745        — everything
grades.mean(axis=1)      # [80. 76.67 91.67]   — each student's average (across subjects)
grades.max(axis=0)       # [100 90 95]         — top score in each subject

axis=1 runs along each row, so it gives one number per student. axis=0 runs down each column, giving one number per subject. Getting this right is most of the battle with NumPy.

Finding where, not just what

argmax and argmin return the index of the biggest or smallest value — handy for "who won?" questions:

averages = grades.mean(axis=1)   # [80. 76.67 91.67]
averages.argmax()                # 2   — student 2 has the highest average

You can summarize arrays in any direction. Next: changing their shape.