
  tteesstt -- tteessttss tthhee nneettwwoorrkk oonn tthhee tteessttiinngg sseett

  UUSSAAGGEE

        test [<num-examples>] [-noreset | -return]

  DDEESSCCRRIIPPTTIIOONN

  This runs the network (forward pass only) on examples in the testing set
  and accumulates the overall error. If there is no testing set, the training
  set is used. The total error is stored in the network's error field. By
  default, a small report is printed to the terminal. The -return flag can be
  used if you instead want the report to be the command's return value.

  In addition to printing the report, the numbers in the report are also
  stored in a global array called Test. You can access the fields in the
  array to easily retrieve the results of the test. The following are the
  available fields:

  * nnuummEExxaammpplleess: The number of examples on which testing was done.
  * nnuummTTiicckkss: The total number of ticks in those examples.
  * ttoottaallEErrrroorr: The total error across all examples.
  * eexxaammpplleeEErrrroorr: The total error per example.
  * ttiicckkEErrrroorr: The total error per tick.
  * ttoottaallCCoosstt: The total unit output cost for the network.
  * ttiicckkCCoosstt: The total unit output cost divided by the number of ticks.
  * eexxaammpplleessCCoorrrreecctt: The number of examples on which the output groups'
    criteria (such as STANDARD_CRIT or MAX_CRIT) were met.
  * ppeerrcceennttCCoorrrreecctt: The percentage of examples for which the criteria were
    met.

  If the network has more than one output group, it prints the error and cost
  values for the network as a whole as well as for each output group. The
  error values for the individual groups are stored in the Test array as
  well. The values for the groups are accessed with the name of the group,
  followed by a . and then the field name. For example, the totalError for
  group output1 is stored in Test(output1.totalError).

  The default behavior is to reset the testing set and run every example in
  it. This is also the behavior if a num-examples value of 0 is given.
  Otherwise, the testing set is not reset and only the next num-examples will
  be processed. This may be necessary if the testing set is in PIPE mode. The
  -noreset flag prevents the network's overall error from being reset to 0.0
  prior to testing.

  EEXXAAMMPPLLEESS

  To test the network, resetting the error before starting:

        lens> test
        Test results on 100 examples, 700 ticks:
                             Network
        Error total:         46.5343
        Error per example:   0.46534
        Error per tick:      0.06648
        Unit cost per tick:  2.33553
        Output unit criterion reached on 92 examples (92.00%)

        lens> puts $Test(tickError)
        0.00664776

  To append the test results to a file:

        lens> set f [open "testResults" "a"]
        lens> puts $f [test -r]
        lens> close $f

  SSEEEE AALLSSOO

  _t_r_a_i_n, _d_o_E_x_a_m_p_l_e, _e_x_a_m_p_l_e_S_e_t_M_o_d_e

  ---------------------------------------------------------------------------
    Last modified: Mon Mar 1 14:43:22 EST 1999

