📄 usage.h
字号:
" examples for Perceptron and Winnow. The first floating point value\n"," represents the distance between the threshold and positive examples.\n"," The second floating point value is optional and represents the\n"," distance between the threshold and negative examples. If the positive\n"," distance is specified but the negative distance isn't, the negative\n"," distance is set equal to the positive distance. See the User's Manual\n"," for more information. Default 0,0.\n","\n"," -s <s | f>\n"," Sets the network type. 's' makes the network sparse, meaning that\n"," features are only linked to a target when they appear in the same\n"," example as that target enough times to become eligible. 'f' makes the\n"," network full, meaning that features become eligible to a target based\n"," on the number of examples they appear in, whether with that target or\n"," not. Default 's'.\n","\n"," -T {string}\n"," This option specifies a file from which test examples are evaluated\n"," after training is finished. When used in conjunction with the -c\n"," paramater, it specifies the file on which the network is tested after\n"," each curve interval. The options listed as training options affecting\n"," testing affect both learning curve testing and the testing done at the\n"," end of training, except for -i which is only applied to the testing\n"," done at the end of training. No default.\n","\n"," -t <+ | ->\n"," '+' enables threshold relative updating for either Winnow or\n"," Perceptron. See the User's Guide for more details. This option has\n"," no effect when '-G +' is also specified. Default '-'.\n","\n"," -u <+ | ->\n"," This option enables or disables first cycle updates. '+' means that\n"," each target's weight vector will be updated during the first cycle of\n"," training. If '-' is specified, features will still become eligible\n"," and linked to targets during the first cycle, but no weights will be\n"," updated. Default '+'.\n","\n"," -V <+ | ->\n"," '+' enables the averaged voted Perceptron algorithm in conjunction\n"," with either Perceptron or Winnow. Two weight vectors are stored in\n"," the network file for each target node; one is the usual weight vector\n"," and the other is the average of all incarnations of the weight vector\n"," during training, weighted by the number of examples they were applied\n"," to. Default '-'.\n","\n"," -v <off | min | med | max>\n"," Specifies verbosity level. Default 'min'.\n","\n"," -w {double}\n"," Specifies a smoothing value for Winnow and Perceptron learners. In\n"," training mode, this option only makes sense when used in conjunction\n"," with the -T parameter. Default 0.\n","\n"," -z <+ | ->\n"," Setting this option to '+' enables \"conventional\" or \"raw\" mode, in\n"," which SNoW adheres to the most traditional definitions of its\n"," algorithms. The network is full instead of sparse, features are\n"," immediately linked to all targets, and activations are not normalized.\n"," Default '-'.\n","\n\n"," Required options for mode -test:\n","\n"," -F {string}\n"," Specifies the name of a file from which the network is read. \n","\n"," -I {string}\n"," Specifies the name of a file from which the testing examples are read.\n","\n"," Other Options for mode -test:\n","\n"," -a <+ | ->\n"," In testing mode, this option makes sense only in conjunction with the\n"," -i parameter. '+' forces all non-discarded features to be written to\n"," be written to the network. The default is '-', when features that\n"," have not yet reached the eligibility threshold are not written to the\n"," network.\n","\n"," -b {double}\n"," Specifies the smoothing parameter to be used in Naive Bayes. Default\n"," 15.0.\n","\n"," -E {string}\n"," Specifies the name of a file in which to write information about\n"," mistakes during testing. No default.\n","\n"," -e <count:{integer} | percent:{double}>\n"," Sets the feature eligibility method. In testing mode,\n"," -e count:{integer} is useful when reading in and initializing the\n"," network. Any feature whose active count (as recorded in the network)\n"," is lower than the eligibility threshold specified will be initialized\n"," as 'pending' (waiting to be linked). Default is 'count:2'.\n","\n"," -f <+ | ->\n"," '+' enables automatic insertion of the \"fixed\" feature into every\n"," example. This feature's weight then acts as a dynamic threshold.\n"," Default '+'.\n","\n"," -G <+ | ->\n"," In testing mode, this parameter only makes sense when used in\n"," conjunction with the -i parameter. When set to '+', this option\n"," enables the Gradient Descent algorithm in conjunction with -P and the\n"," Exponentiated Gradient Descent algorithm with -W. This parameter has\n"," no effect when used with -B. Also, it cannot be used in conjunction\n"," with either '-O +' or '-t +'. Default '-'.\n","\n"," -g <+ | ->[,<+ | ->]\n"," The first argument of this parameter only has effect on -train mode.\n","\n"," If conjunctions were generated during training, SNoW will\n"," automatically generate conjunctions in the testing examples before\n"," presenting them to the network for classification also. Users also\n"," have the option of writing the new examples to disk by specifying a\n"," second argument. If '-g +,+' is specified, input examples will be\n"," written to disk with conjunctions added. They will be output into a\n"," file whose name is the original filename concatenated with\n"," '.conjunctions'. The default is '-g <unset>,-', where no examples are\n"," written to disk.\n","\n"," -i <+ | ->\n"," This option specifies whether incremental learning should be used.\n"," '+' presents each example to the network for training. The resulting\n"," network is written at the end of testing with .new appended to its\n"," filename. Default '-'.\n","\n"," -L {long}\n"," Specifies the limit to the number of targets printed with the -o\n"," parameter. This option makes sense only when used in conjunction with\n"," the -o parameter. Default ULONG_MAX.\n","\n"," -l <+ | ->\n"," This option specifies whether test examples are labeled or not.\n"," Default '+'.\n","\n"," -m <+ | ->\n"," This option specifies whether to train with multiple labels. '+'\n"," means that a given target will not treat other targets' IDs as\n"," features when they are encountered in an example. '-' means that a\n"," given target treats all IDs as features except for its own ID. When\n"," making predictions, having multiple targets set to '+' will cause all\n"," examples that contain the predicted target anywhere in the example to\n"," be counted correct. If set to '-', the predicted target must be the\n"," first target to appear in the example for the example to be counted\n"," correct. This option will have an affect on the smoothing of\n"," previously unseen features during testing (see -b and -w). If set to\n"," '-', a given target will treat other target IDs as features which can\n"," potentially be smoothed. Default '+'.\n","\n"," -O <+ | ->[,<+ | ->]\n"," In testing mode, this option only makes sense when used in conjunction\n"," with the -i parameter. '+' enables ordered targets (a.k.a. constraint\n"," classification) mode. '+' in the second argument is used to enable a\n"," more conservative version of the algorithm. If the first argument has\n"," been set to '+' and the second is unspecified, it is the same as\n"," specifying '+,-'. Default '-,-'.\n","\n"," -o < accuracy | winners | softmax | allpredictions | allactivations |\n"," allboth >\n"," Specifies which output mode to use when reporting test results.\n"," Default 'accuracy'.\n"," accuracy:\n"," This mode requires labeled examples. Output is given as a\n"," percentage accuracy for all of the test examples.\n"," winners:\n"," This mode outputs the targetID with the highest activation for each\n"," example.\n"," softmax:\n"," This mode outputs, for each example, a list of all targets and their\n"," corresponding softmax normalized activations.\n"," allpredictions:\n"," This mode outputs, for each example, a list of all targets and their\n"," predictions (0 for false, 1 for true). The target with the highest\n"," activation will be predicted as true, all others false.\n"," allactivations:\n"," This mode outputs, for each example, a list of all targets and their\n"," activations.\n"," allboth:\n"," This mode outputs, for each example, a list of all targets and both\n"," their activations and predictions.\n","\n"," -p {double}\n"," Specifies the prediction threshold. SNoW will not make a prediction\n"," if the activation of the most activated target minus the activation of\n"," the second most activated target is less than or equal to the\n"," prediction threshold. Default -1.\n","\n"," -R {string}\n"," Specifies the name of a file in which to write the testing results.\n"," Default STDOUT.\n","\n"," -S {double}[,{double}]\n"," In testing mode, this option only makes sense in conjunction with the\n"," -i parameter. Sets the thickness of the separator between positive\n"," and negative examples for Perceptron and Winnow. Default 0,0.\n","\n"," -s <s | f>\n"," In testing mode, this option only makes sense in conjunction with the\n"," -i parameter. Sets the network type. 's' makes the network sparse,\n"," meaning that features are only linked to a target when they appear in\n"," the same example as that target enough times to become eligible. 'f'\n"," makes the network full, meaning that features become eligible to a\n"," target based on the number of examples they appear in, whether with\n"," that target or not. Default 's'.\n","\n"," -t <+ | ->\n"," In testing mode, this option only makes sense in conjunction with the\n"," -i parameter. '+' enables threshold relative updating for either\n"," Winnow or Perceptron. See the User's Guide for more details. This\n"," message has no effect when '-G +' is also specified. Default '-'.\n","\n"," -V <+ | ->\n"," '+' enables averaged voted Perceptron testing in conjunction with\n"," either Perceptron or Winnow. Two weight vectors are stored in the\n"," network file for each target node when the network is trained with\n"," '-V +'; one is the usual weight vector and the other is the average\n"," of all incarnations of the weight vector during training, weighted by\n"," the number of examples they were applied to. In testing, this flag\n"," directs SNoW to use the averaged weight vector when classifying\n"," testing examples. It also enables the averaged voted Perceptron\n"," algorithm in case '-i +' is also set. Default '-'.\n","\n"," -v <off | min | med | max>\n"," Specifies verbosity level. Default 'min'.\n","\n"," -w {double}\n"," Specifies a smoothing value for Winnow and Perceptron learners.\n"," Default 0.\n","\n"," -z <+ | ->\n"," Setting this option to '+' enables \"conventional\" or \"raw\" mode, in\n"," which SNoW adheres to the most traditional definitions of its\n"," algorithms. The network is full instead of sparse, features are\n"," immediately linked to all targets, and activations are not normalized.\n"," Default '-'.\n","\n\n"," Required options for mode -interactive:\n","\n"," -F {string}\n"," Specifies the name of a file to which the resulting network is written\n"," after training. No default.\n","\n"," -I {string}\n"," Specifies the name of a file from which training examples are read.\n"," No default.\n","\n"," Architecture definition options for mode -interactive: \n","\n"," -A {string}\n"," Specifies the name of a file from which to read the algorithm\n"," definition. No default.\n","\n"," -P <learning_rate,threshold,default_weight:targets>\n"," Adds a Perceptron algorithm to the network for the given targets.\n","\n"," -W <alpha,beta,threshold,default_weight:targets>\n"," Adds a Winnow algorithm to the network for the given targets.\n","\n"," Other options for mode -interactive:\n",⌨️ 快捷键说明
复制代码
Ctrl + C
搜索代码
Ctrl + F
全屏模式
F11
切换主题
Ctrl + Shift + D
显示快捷键
?
增大字号
Ctrl + =
减小字号
Ctrl + -