PERF Get rid of intermediate array in binary_log_loss

[alexhenrie]

This decreases by about 10% the time required to train an example neural network from binary data. It also saves memory!

Test program:

from sklearn.datasets import load_digits
from sklearn.neural_network import MLPClassifier
from neurtu import delayed, Benchmark

digits = load_digits(return_X_y=True)
X = digits[0][:,:50].astype(bool).astype(int)
y = digits[0][:,50].astype(bool).astype(int)

clf = MLPClassifier(solver='lbfgs', alpha=1e-5,
                    hidden_layer_sizes=(50, 20), random_state=1, max_iter=1000)

train = delayed(clf).fit(X, y)
print(Benchmark(wall_time=True, cpu_time=True, repeat=5)(train))

Before:

      wall_time  cpu_time                                                                                                     
mean   8.084272  8.149094
max    8.106030  8.485355
std    0.016399  0.188559

After:

      wall_time  cpu_time                                                                                                     
mean   7.310994  7.283616
max    7.332577  7.296908
std    0.014912  0.014172

[rth]

Nice, thank you @alexhenrie !

[thomasjpfan]

The work you have done on the neural_network module has been great! @alexhenrie

LGTM!