There was an issue with the min cost flow input. about k-means-constrained

[joshlk]

Hi, thanks for taking the time and filing a issue. Can you please provide the full error as I think you have forgot to include it? Could you also provide me with the sklearn, scipy and numpy version. And what system are you using? Mac/Linux/Windows?

[zjorgensenbits]

Thanks for the quick response, and sorry for the omission. The exception is getting thrown by this check on line 483 of k_means_contrained_.py:

if min_cost_flow.Solve() != min_cost_flow.OPTIMAL:
   raise Exception('There was an issue with the min cost flow input.')

Digging into it deeper, I see that min_cost_flow.Solve() is returning 3, while min_cost_flow.OPTIMAL is 1.

Versions:
numpy==1.13.3
scikit_learn==0.24.1
k-means-constrained==0.5.1

OS: Ubuntu 18.04.5

[joshlk]

Could you also please provide a minimal working example that produces the error

[zjorgensenbits]

from k_means_constrained import KMeansConstrained
import numpy as np

X = np.array([[11.43817265911908, 3.0796017788793604], [8.656592391707292, -0.5358447753622294], [7.302104968579831, -2.2195335805441103], [7.1817782508003924, 5.142237391259907], [6.672130682774286, 1.9154397778535366], [6.335350488151739, -5.14734393191299], [-4.467719064297461, 2.198353204833053], [6.259064200368242, 4.7330446470564445], [10.769348531911556, 0.7475931768575942], [12.555003395037621, -0.8724602182508533], [10.9665526970981, -4.4246780352423]])

clf = KMeansConstrained(n_clusters=2, size_min=2, size_max=5, random_state=0)
clf.fit_predict(X)

---

Traceback (most recent call last):
File "/home/tatooine/.local/lib/python3.6/site-packages/IPython/core/interactiveshell.py", line 3343, in run_code
exec(code_obj, self.user_global_ns, self.user_ns)
File "", line 5, in
clf.fit_predict(X)
File "/home/tatooine/.local/lib/python3.6/site-packages/k_means_constrained/k_means_constrained_.py", line 730, in fit_predict
return self.fit(X).labels_
File "/home/tatooine/.local/lib/python3.6/site-packages/k_means_constrained/k_means_constrained_.py", line 642, in fit
return_n_iter=True)
File "/home/tatooine/.local/lib/python3.6/site-packages/k_means_constrained/k_means_constrained_.py", line 180, in k_means_constrained
x_squared_norms=x_squared_norms, random_state=random_state)
File "/home/tatooine/.local/lib/python3.6/site-packages/k_means_constrained/k_means_constrained_.py", line 328, in kmeans_constrained_single
labels_constrained(X, centers, size_min, size_max, distances=distances)
File "/home/tatooine/.local/lib/python3.6/site-packages/k_means_constrained/k_means_constrained.py", line 399, in labels_constrained
labels = solve_min_cost_flow_graph(edges, costs, capacities, supplies, n_C, n_X)
File "/home/tatooine/.local/lib/python3.6/site-packages/k_means_constrained/k_means_constrained.py", line 484, in solve_min_cost_flow_graph
raise Exception('There was an issue with the min cost flow input.')
Exception: There was an issue with the min cost flow input.

[joshlk]

Hi @zjorgensenbits,

The constraints you have provided make the problem unsolvable i.e. you have 11 data points and said you want 2 clusters with max size 5. There is no way of solving this as 2*5=10<11.

However, the user input should be validated before trying to solve the solve_min_cost_flow_graph - I will add to the code 😃 thanks.

Josh

[zjorgensenbits]

doh! you're right--I'm an idiot. I misinterpreted "min_size" and "max_size" as the min and max number of clusters...
I'll go ahead and close this. Thanks for catching my error!