**What it does**

This tool takes a label file and a prediction file as input and
computes several measures of accuracy of the predictions.  In
particular, the areas under the Receiver-Operator-Curve (ROC) and the
Precision-Recall-Curve (PRC) are assessed.  Both measures are close to
1.0 for nearly perfect classification; for random guessing they are
close to 0.5 (ROC) and 0.0 (PRC), respectively.

Typical values for the area under the ROC curve are between 0.98 and
0.998 for splice sites, between 0.85 and 0.95 for transcription start
and cleavage sites, and around 0.95 for translation initation sites
and stop codons.

------

**Output**

This tool generates a brief accuracy report, stored as a text file for
human inspection.

--------

**Example**

The evaluation of acceptor splice site predictions may look like this::

  Evaluation for "acc" predictions on 981051 positions
   * Area under ROC curve: 0.986
   * Area under PRC curve: 0.854

------

**About formats**

**SPF format** Signal Prediction format was designed for providing information such as labels or predictions for specific genomic locations (with strand information). It has exacty six required fields::

    1. chrom - The name of the chromosome (e.g. chr1, chrY_random).
    2. signalName - possible choices include tss, tis, acc, don, cdsStop, cleave.
    3. scoreName - possible choices include label, output, Conf, Conf_Cum
    4. chromPos - The position in the chromosome. (The first base in a chromosome is numbered 1.)
    5. strand - Defines the strand - either '+' or '-'.
    6. score - The score between -infinity and infinity. If scoreName is 'label', then the score should be either -1 or 1.

