Skip to content
Advertisement

Numpy find maximum tuple in array of windows

I’m starting our with list of tuples (each tuple is an (X,Y)). My end result is I want to find the maximum Y-value within EACH window/bin of length 4 using numpy.

# List of tuples
[(0.05807200929152149, 9.9720125), (0.34843205574912894, 1.1142874), (0.6387921022067363, 2.0234027), (0.9291521486643438, 1.4435122), (1.207897793263647, 2.3677678), (1.4982578397212543, 1.9457655), (1.7886178861788617, 2.8343441), (2.078977932636469, 5.7816567), ...]

# Convert to numpy array
dt = np.dtype('float,float')
arr = np.asarray(listTuples, dt)
# [(0.05807201, 9.97201252) (0.34843206, 1.11428738)
#  (0.6387921 , 2.02340269) (0.92915215, 1.4435122 )
#  (1.20789779, 2.36776781) (1.49825784, 1.9457655 )
#  (1.78861789, 2.83434415) (2.07897793, 5.78165674)
#  (2.36933798, 3.14842606) ...]

#Create windows/blocks of 4 elements
arr = arr.reshape(-1,4)
# [[(0.05807201, 9.97201252) (0.34843206, 1.11428738)
#   (0.6387921 , 2.02340269) (0.92915215, 1.4435122 )]
#  [(1.20789779, 2.36776781) (1.49825784, 1.9457655 )
#   (1.78861789, 2.83434415) (2.07897793, 5.78165674)]
#  [(2.36933798, 3.14842606) (2.95005807, 2.10357308)
#   (3.24041812, 1.15985966) (3.51916376, 2.03056955)]...]

print(arr.max(axis=1)) <-- ERROR HERE
print(max(arr,key=lambda x:x[1])) <-- ERROR, tried this too but doesn't work

Expected output I want from each window/blocks using max y-value is below. Alternatively format could be regular list of tuples, doesn’t strictly need to be numpy array:

[[(0.05807201, 9.97201252)]
[(2.07897793, 5.78165674)]
[(2.36933798, 3.14842606)]...]
OR other format:
[(0.05807201, 9.97201252),(2.07897793, 5.78165674),(2.36933798, 3.14842606)]...]

Advertisement

Answer

Here is a vectorized (and probably fastest) approach:

arr = np.asarray(listTuples, np.dtype('float,float'))
idx = arr.view(('float',2))[:,1].reshape(-1,4).argmax(1)
arr = arr.reshape(-1,4)[np.arange(len(idx)),idx]
#[(0.05807201, 9.9720125) (2.07897793, 5.7816567) ...]

You basically use a array (non-structured) version of your structured array by view and find argmax of Y along axis=1. Then use those indices to filter the tuples from your original array arr.

User contributions licensed under: CC BY-SA
8 People found this is helpful
Advertisement