Trains/updates the MLP

status = model.train(samples, responses)
status = model.train(csvFilename, [])
[...] = model.train(..., 'OptionName', optionValue, ...)

Input

• samples matrix of training samples. It should be floating-point type. By default, each row represents a sample (see the Layout option).
• responses Floating-point matrix of the corresponding output vectors, one vector per row.
• csvFilename The input CSV file name from which to load dataset. In this variant, you should set the second argument to an empty array.

Output

• status Success flag.

Options

• Data Training data options, specified as a cell array of key/value pairs of the form {'key',val, ...}. See below.
• Flags The optional training flags, model-dependent. For convenience, you can set the individual flag options below, instead of directly setting bits here. default 0
• UpdateWeights Algorithm updates the network weights, rather than computes them from scratch. In the latter case the weights are initialized using the Nguyen-Widrow algorithm. default false.
• NoInputScale Algorithm does not normalize the input vectors. If this flag is not set, the training algorithm normalizes each input feature independently, shifting its mean value to 0 and making the standard deviation equal to 1. If the network is assumed to be updated frequently, the new training data could be much different from original one. In this case, you should take care of proper normalization. default false.
• NoOutputScale Algorithm does not normalize the output vectors. If the flag is not set, the training algorithm normalizes each output feature independently, by transforming it to the certain range depending on the used activation function. default false.

Options for Data (first variant with samples and reponses)

• Layout Sample types. Default 'Row'. One of:
  • Row each training sample is a row of samples.
  • Col each training sample occupies a column of samples.
• VarIdx vector specifying which variables to use for training. It can be an integer vector (int32) containing 0-based variable indices or logical vector (uint8 or logical) containing a mask of active variables. Not set by default, which uses all variables in the input data.
• SampleIdx vector specifying which samples to use for training. It can be an integer vector (int32) containing 0-based sample indices or logical vector (uint8 or logical) containing a mask of training samples of interest. Not set by default, which uses all samples in the input data.
• SampleWeights optional floating-point vector with weights for each sample. Some samples may be more important than others for training. You may want to raise the weight of certain classes to find the right balance between hit-rate and false-alarm rate, and so on. Not set by default, which effectively assigns an equal weight of 1 for all samples.
• VarType optional vector of type uint8 and size <num_of_vars_in_samples> + <num_of_vars_in_responses>, containing types of each input and output variable. By default considers all variables as numerical (both input and output variables). In case there is only one output variable of integer type, it is considered categorical. You can also specify a cell-array of strings (or as one string of single characters, e.g 'NNNC'). Possible values:
  • Numerical, N same as 'Ordered'
  • Ordered, O ordered variables
  • Categorical, C categorical variables
• MissingMask Indicator mask for missing observation (not currently implemented). Not set by default
• TrainTestSplitCount divides the dataset into train/test sets, by specifying number of samples to use for the test set. By default all samples are used for the training set.
• TrainTestSplitRatio divides the dataset into train/test sets, by specifying ratio of samples to use for the test set. By default all samples are used for the training set.
• TrainTestSplitShuffle when splitting dataset into train/test sets, specify whether to shuffle the samples. Otherwise samples are assigned sequentially (first train then test). default true

Options for Data (second variant for loading CSV file)

• HeaderLineCount The number of lines in the beginning to skip; besides the header, the function also skips empty lines and lines staring with '#'. default 1
• ResponseStartIdx Index of the first output variable. If -1, the function considers the last variable as the response. If the dataset only contains input variables and no responses, use ResponseStartIdx = -2 and ResponseEndIdx = 0, then the output variables vector will just contain zeros. default -1
• ResponseEndIdx Index of the last output variable + 1. If -1, then there is single response variable at ResponseStartIdx. default -1
• VarTypeSpec The optional text string that specifies the variables' types. It has the format ord[n1-n2,n3,n4-n5,...]cat[n6,n7-n8,...]. That is, variables from n1 to n2 (inclusive range), n3, n4 to n5 ... are considered ordered and n6, n7 to n8 ... are considered as categorical. The range [n1..n2] + [n3] + [n4..n5] + ... + [n6] + [n7..n8] should cover all the variables. If VarTypeSpec is not specified, then algorithm uses the following rules:
  • all input variables are considered ordered by default. If some column contains has non- numerical values, e.g. 'apple', 'pear', 'apple', 'apple', 'mango', the corresponding variable is considered categorical.
  • if there are several output variables, they are all considered as ordered. Errors are reported when non-numerical values are used.
  • if there is a single output variable, then if its values are non-numerical or are all integers, then it's considered categorical. Otherwise, it's considered ordered.
• Delimiter The character used to separate values in each line. default ','
• Missing The character used to specify missing measurements. It should not be a digit. Although it's a non-numerical value, it surely does not affect the decision of whether the variable ordered or categorical. default '?'
• TrainTestSplitCount same as above.
• TrainTestSplitRatio same as above.
• TrainTestSplitShuffle same as above.

This method applies the specified training algorithm to computing/adjusting the network weights.

The RPROP training algorithm is parallelized with the TBB library.

If you are using the default 'Sigmoid' activation function then the output should be in the range [-1,1], instead of [0,1], for optimal results.