DecisionTreeModel

DecisionTreeModel = class DecisionTreeModel(Modifiable, Dynamic, LinearModel)

DecisionTreeModel(ndim=1, depth=0, split=0.5, kdim=0, itypes=[0], modifiable=True, dynamic=True, code=None, growPrior=None, copy=None, **kwargs) A DecisionTree Model (DTM) is mostly defined on multiple input dimensions (axes). It splits the data in 2 parts, according low and high values on a certain input axis. The splitting can continue along other axes. The axes can contain float values, categorials (int) or booleans. The float axes can have Nans when data are unkown. Each category set one bit in an integer. The unknown category is a category by itself. Booleans that contain unknowns should be coded as categorial. f( x:p ) = DTM \left => DTM (or None) ite => DTM (or None) \dimension \split or \mask The tree is searched left to right. The parameters are all initialized at 0.0 Examples -------- >>> dtm = DecisionTreeModel( ) >>> print( dtm ) DecisionTree: with 0 components and 1 parameters Attributes ---------- left : None or DTM a None the tree stops otherwise there is a new split. rite : None or DTM a None the tree stops otherwise there is a new split. dimension : int split according to data values on this axes itype : char. Either 'f' (float), 'i' (integer), or 'b' (boolean) characterizes the input dimension as float, integer or boolean split : float between 0 and 1 (for float dimension) to the left normalized values < split; to the rite normvalues > split. Unknown values (NaNs to to the smallest faction). nsplit : int number of calls to random.rand() to be averaged. Prior on "split" mask : float between 0 and 1 (for float dimension) to the left normalized values < split; to the rite normvalues > split. Unknown values (NaNs to to the smallest faction. Attributes from Modifiable -------------------------- modifiable Attributes from Dynamic ----------------------- dynamic Attributes from Model --------------------- npchain, parameters, stdevs, xUnit, yUnit Attributes from FixedModel -------------------------- npmax, fixed, parlist, mlist Attributes from BaseModel -------------------------- npbase, ndim, priors, posIndex, nonZero, tiny, deltaP, parNames

Method resolution order:: DecisionTreeModel; Modifiable; Dynamic; LinearModel; Model; FixedModel; BaseModel; builtins.object

Constructor:

DecisionTreeModel( ndim=1, depth=0, split=0.5, kdim=0, itypes=[0],
modifiable=True, dynamic=True, code=None, growPrior=None, copy=None,
**kwargs): DecisionTree model. The DTM standardly has a UniformPrior for all parameters, with limits [0,1] Parameters ---------- ndim : int number of input dimensions depth : int depth of the tree kdim : None or list of ints input channel to be splitted. itypes : [list of] int indicating the type of input: 0 for float, 1 for boolean, >1 for categorial The last number is repeated for remaining inputs split : None or float or list of floats fraction (0<s<1) of the input kdim that falls on either side. growPrior : None or Prior Governing the growth modifiable : bool (True) Will the model modify dimension and/or split/mask dynamic : bool (True) Will the model grow/shrink

Methods defined here:

baseDerivative( xdata, params ): Return the derivative df/dx at each xdata (=x). Parameters ---------- xdata : array_like values at which to calculate the result params : array_like values for the parameters.

baseName(): Returns a string representation of the model.

baseParameterUnit( k ): Return the unit of the indicated parameter. Parameters --------- k : int parameter number.

basePartial( xdata, params, parlist=None ): Returns the partials at the input value. Parameters ---------- xdata : array_like values at which to calculate the partials params : array_like values for the parameters. parlist : array_like list of indices active parameters (or None for all)

baseResult( xdata, params ): Returns the result of the model function. Parameters ---------- xdata : array_like values at which to calculate the result params : array_like values for the parameters.

check(): Find a branch in the tree, returning the branch. Parameter --------- kbranch : int index of the branch to be found

copy(): Copy method.

count(): Return number of leafs and branches.

countBranch(): Return number of leafs.

countLeaf(): Return number of leafs.

decode( code, kdim, splim, kbr ): Resurrect the DTM from the code generated by encode(). For internal use in the Constructor. Parameters ---------- code : list of (list or 1 or 2), encoding the structure of DTM dims : list of dimensions at the branches splim : list of split/mask values at the branches kbr : int counter (start at 0)

encode(): Make a code tuple to be used by the constructor to resurrect the DTM The tuple consists of code : list of (list or 1 or 2), encoding the structure of DTM dims : list of dimensions at the branches splim : list of split/mask values at the branches

findBranch( kbranch ): Find a branch in the tree, returning the branch. Parameter --------- kbranch : int index of the branch to be found

findLeaf( kleaf ): Find a leaf in the tree, returning the leaf. Parameter --------- kleaf : int index of the leaf to be found

findRoot(): Return the root of the tree.

fullName( ids=False ): Returns a string representation of the model. Parameters ---------- ids : bool if True give the pointers of the links too.

grow( offset=0, rng=None, location=0, split=0.5, kdim=0): Increase the the number of components by 1 (if allowed by maxComp) Parameters ---------- offset : int offset in the parameter list (pertaining to earlier models in the chain) rng : Random Number Generator to obtain random values for items below. location : int location where the new dtm-leaf should be inserted kdim : int (<self.ndim) dimension to split split : float (0<split<1) relative cut on this dimension Return ------ bool : succes

isLeaf(): Return true if is a leaf

partitionList( xdata, plist ): Partition the xdata in plist over 2 new lists according to the DTM-branch. Paramaters ---------- xdata : array-like the xdata plist : array of ints indices in xdata Return ------ pl1, pl2 : 2 lists together containing all indices in plist

recursiveName( name, indent, kpar, ids=False ): For internal use only

recursiveOrder( xdata, plist ): For internal use only

recursivePartial( xdata, kpar, plist, part ): For internal use only

recursiveResult( xdata, params, kpar, plist, res ): For internal use only

setSplitOrMask( itype, split ): For internal use only

shrink( offset=0, rng=None, location=0): Decrease the the number of componenets by 1 (if allowed by minComp) Remove an arbitrary item. Parameters ---------- offset : int location where the new params should be inserted Return ------ bool : succes

sortXdata( xdata ): Reorder the xdata according to the parameter ordering Parameters ---------- xdata : array_like values at which to calculate the partials

vary( rng=None, location=0, split=0.5, kdim=0): Vary the model structure by changing kdim and/or split at location Parameters ---------- rng : Random Number Generator to obtain random values for items below. location : int location where the dtm-branch should be changed kdim : int (<self.ndim) dimension to split split : float (0<split<1) relative cut on this dimension Return ------ bool : succes

walk(): Iterate tree in pre-order depth-first search order Found this piece of code on the internet. Fairly obscure.

Methods inherited from Modifiable:

isModifiable()

Methods inherited from Dynamic:

Methods inherited from Model:

Overloaded operators and aliases

Other methods

Methods inherited from FixedModel:

Methods inherited from BaseModel: