Confusion matrix util modified

xbob/boosting/util/confusion.py

 import numpy
 import matplotlib.pyplot 
 
+def confusion_matrix(expected, predicted):
+  """Transforms the prediction list into a confusion matrix
+  
+  This method takes lists of expected and predicted classes and returns a
+  confusion matrix, which represents the percentage of classified examples in
+  each combination of "expected class" of samples and "predicated class" of the
+  same samples.
+
+  Keyword parameters:
+
+  expected (numpy.ndarray, 1D)
+    The ground-thruth
+
+  predicted (numpy.ndarray, 1D)
+    The predicted classes with your neural network
+
+  You must combine these scores column wise and determine what are the
+  annotated rates (below) for each of the column entries, returning the
+  following 2D numpy.ndarray::
+
+    [ TP0    / N0    FP1(0) / N1    FP2(0) / N2 ... ]
+    [ FP0(1) / N0    TP1    / N1    FP2(1) / N2 ... ]
+    [ FP0(2) / N0    FP1(2) / N1    TP2    / N2 ... ]
+    [     ...              ...            ...       ]
+    [ FP0(9) / N0    FP1(9) / N1    TP9    / N9 ... ]
+
+  Where:
+
+  TPx / Nx
+    True Positive Rate for class ``x``
+
+  FPx(y) / Nz
+    Rate of False Positives for class ``y``, from class ``x``. That is,
+    elements from class ``x`` that have been **incorrectly** classified as
+    ``y``.
+  """
+
+  retval = numpy.zeros((10,10), dtype=float)
+
+  for k in range(10):
+    pred_k = predicted[expected==k] # predictions that are supposed to be 'k'
+    retval[:,k] = numpy.array([len(pred_k[pred_k==p]) for p in range(10)])
+    retval[:,k] /= len(pred_k)
+
+  return retval
+
+
 
 
 def display_cm(cm, title_str = "Confusion Matrix"):