gist-score-svm

Description: Compute various statistics from svm output files.

Usage: gist-score-svm [options] <weights>

Inputs:

• weights - an SVM weights file produced by compute-weights

Output:

Number of training examples: 985 (477 positive, 508 negative)
Number of support vectors: 498 (217 positive, 281 negative) 50.6%
Training results: FP=0 FN=0 TP=477 TN=508
Training ROC: 1.00000
Test results: FP=103 FN=107 TP=370 TN=405
Test ROC: 0.85417

• The first line of output reports the number of examples in the training set.
• The second line reports the number and percentage of those examples that are support vectors (i.e., for which the weight is non-zero).
• The third line will only be printed if the weights file specifies constraints on the magnitudes of the support vector weights. This line reports the number and percentage of support vectors whose weights reached the set bounds. Note that the percentage reported on this line is with respect to the entire data set, not the set of support vectors.
• The fourth line reports the number of false positive, false negative, true positive and true negative classifications of the training set.
• The fifth line reports the ROC score, which is the normalized area under a curve that plots true positives as a function of false positives for varying decision thresholds. Perfect classification corresponds to an ROC of 1.0.
• The sixth and seventh lines report the same values as the fourth and fifth lines, but with respect to the test set. The test set statistics may computed with respect to hold-one-out predictions if the weights file contains them. Alternatively, the test set may be supplied as additional files, using the -test option. If no test set is given, these two lines will not appear in the output.

Options:

• -test <truth> <predictions> - Compute the total number of errors with respect to an external test set.
• -rdb - Assume that the input files contain RDB format lines.

Calls: none