public (Tree.Ensemble, Parameters) GetModel()
{
Tree.Ensemble res = new Tree.Ensemble();
string modelString = GetModelString();
string[] lines = modelString.Split('\n');
var prms = new Dictionary <string, string>();
int i = 0;
for (; i < lines.Length;)
{
if (lines[i].StartsWith("Tree="))
{
Dictionary <string, string> kvPairs = new Dictionary <string, string>();
++i;
while (!lines[i].StartsWith("Tree=") && lines[i].Trim().Length != 0)
{
string[] kv = lines[i].Split('=');
if (kv.Length != 2)
{
throw new FormatException();
}
kvPairs[kv[0].Trim()] = kv[1].Trim();
++i;
}
int numLeaves = int.Parse(kvPairs["num_leaves"]);
int numCat = int.Parse(kvPairs["num_cat"]);
if (numLeaves > 1)
{
var leftChild = Str2IntArray(kvPairs["left_child"], ' ');
var rightChild = Str2IntArray(kvPairs["right_child"], ' ');
var splitFeature = Str2IntArray(kvPairs["split_feature"], ' ');
var threshold = Str2DoubleArray(kvPairs["threshold"], ' ');
var splitGain = Str2DoubleArray(kvPairs["split_gain"], ' ');
var leafOutput = Str2DoubleArray(kvPairs["leaf_value"], ' ');
var decisionType = Str2UIntArray(kvPairs["decision_type"], ' ');
for (var j = 0; j < threshold.Length; j++)
{
// See 'AvoidInf' in lightgbm source
var t = threshold[j];
if (t == 1e300)
{
threshold[j] = double.PositiveInfinity;
}
else if (t == -1e300)
{
threshold[j] = double.NegativeInfinity;
}
}
var defaultValue = GetDefaultValue(threshold, decisionType);
var categoricalSplit = new bool[numLeaves - 1];
var catBoundaries = Array.Empty <int>();
var catThresholds = Array.Empty <uint>();
if (numCat > 0)
{
catBoundaries = Str2IntArray(kvPairs["cat_boundaries"], ' ');
catThresholds = Str2UIntArray(kvPairs["cat_threshold"], ' ');
for (int node = 0; node < numLeaves - 1; ++node)
{
categoricalSplit[node] = GetIsCategoricalSplit(decisionType[node]);
}
}
var tree = Tree.RegressionTree.Create(
numLeaves,
splitFeature,
splitGain,
threshold,
defaultValue,
leftChild,
rightChild,
leafOutput,
catBoundaries,
catThresholds,
categoricalSplit);
res.AddTree(tree);
}
else
{
// always need to add tree, otherwise multiclass will be wrong
var leafOutput = Str2DoubleArray(kvPairs["leaf_value"], ' ');
var tree = Tree.RegressionTree.Create(
2,
new int[] { 0 },
new double[] { 0 },
new double[] { 0 },
new double[] { 0 },
new int[] { -1 },
new int[] { -2 },
new double[] { leafOutput[0], leafOutput[0] },
new int[] { },
new uint[] { },
new bool[] { false });
res.AddTree(tree);
}
}
else
{
// [objective: binary]
if (lines[i].StartsWith("["))
{
var bits = lines[i].Split(new char[] { '[', ']', ' ', ':' }, StringSplitOptions.RemoveEmptyEntries);
if (bits.Length == 2) // ignores, e.g. [data: ]
{
prms.Add(bits[0], bits[1]);
}
}
++i;
}
}
// extract parameters
var p = new Parameters {
Common = _helperCommon.FromParameters(prms),
Dataset = _helperDataset.FromParameters(prms),
Objective = _helperObjective.FromParameters(prms),
Learning = _helperLearning.FromParameters(prms)
};
return(res, p);
}
public (Tree.Ensemble, Parameters) GetModel()
{
Tree.Ensemble res = new Tree.Ensemble();
string modelString = GetModelString();
string[] lines = modelString.Split('\n');
var prms = new Dictionary<string, string>();
var delimiters = new char[] { ' ' };
int i = 0;
for (; i < lines.Length;)
{
if (lines[i].StartsWith("Tree="))
{
Dictionary<string, string> kvPairs = new Dictionary<string, string>();
++i;
while (!lines[i].StartsWith("Tree=") && lines[i].Trim().Length != 0)
{
string[] kv = lines[i].Split('=');
if (kv.Length != 2) throw new FormatException();
kvPairs[kv[0].Trim()] = kv[1].Trim();
++i;
}
int numLeaves = int.Parse(kvPairs["num_leaves"]);
int numCat = int.Parse(kvPairs["num_cat"]);
var leftChild = Str2IntArray(kvPairs["left_child"], delimiters);
var rightChild = Str2IntArray(kvPairs["right_child"], delimiters);
var splitFeature = Str2IntArray(kvPairs["split_feature"], delimiters);
var threshold = Str2DoubleArray(kvPairs["threshold"], delimiters);
var splitGain = Str2DoubleArray(kvPairs["split_gain"], delimiters);
var leafOutput = Str2DoubleArray(kvPairs["leaf_value"], delimiters);
var decisionType = Str2UIntArray(kvPairs["decision_type"], delimiters);
for (var j = 0; j < threshold.Length; j++)
{
// See 'AvoidInf' in lightgbm source
var t = threshold[j];
if (t == 1e300)
threshold[j] = double.PositiveInfinity;
else if (t == -1e300)
threshold[j] = double.NegativeInfinity;
}
var defaultValue = GetDefaultValue(threshold, decisionType);
var categoricalSplit = new bool[numLeaves - 1];
var catBoundaries = Array.Empty<int>();
var catThresholds = Array.Empty<uint>();
if (numCat > 0)
{
catBoundaries = Str2IntArray(kvPairs["cat_boundaries"], delimiters);
catThresholds = Str2UIntArray(kvPairs["cat_threshold"], delimiters);
for (int node = 0; node < numLeaves - 1; ++node)
{
categoricalSplit[node] = GetIsCategoricalSplit(decisionType[node]);
}
}
double[] leafConst = null;
int[][] leafFeaturesUnpacked = null;
double[][] leafCoeffUnpacked = null;
var isLinear = Int32.Parse(kvPairs["is_linear"]) > 0;
if (isLinear)
{
leafConst = Str2DoubleArray(kvPairs["leaf_const"], delimiters);
var numFeatures = Str2IntArray(kvPairs["num_features"], delimiters);
var leafFeatures = Str2IntArray(kvPairs["leaf_features"], delimiters);
var leafCoeff = Str2DoubleArray(kvPairs["leaf_coeff"], delimiters);
leafFeaturesUnpacked = new int[numFeatures.Length][];
leafCoeffUnpacked = new double[numFeatures.Length][];
var idx = 0;
for (var j=0; j < numFeatures.Length; j++)
{
var len = numFeatures[j];
leafFeaturesUnpacked[j] = new int[len];
leafCoeffUnpacked[j] = new double[len];
for (var k = 0; k < len; k++)
{
leafFeaturesUnpacked[j][k] = leafFeatures[idx];
leafCoeffUnpacked[j][k] = leafCoeff[idx];
idx++;
}
}
if (idx != leafFeatures.Length)
throw new Exception("Failed to parse leaf features");
}
var tree = Tree.RegressionTree.Create(
numLeaves,
splitFeature,
splitGain,
threshold,
defaultValue,
leftChild,
rightChild,
leafOutput,
catBoundaries,
catThresholds,
categoricalSplit,
isLinear,
leafConst,
leafFeaturesUnpacked,
leafCoeffUnpacked);
res.AddTree(tree);
}
else
{
// [objective: binary]
if (lines[i].StartsWith("["))
{
var bits = lines[i].Split(new char[] { '[', ']', ' ', ':' }, StringSplitOptions.RemoveEmptyEntries);
if (bits.Length == 2) // ignores, e.g. [data: ]
prms.Add(bits[0], bits[1]);
}
++i;
}
}
// extract parameters
var p = new Parameters {
Common = _helperCommon.FromParameters(prms),
Dataset = _helperDataset.FromParameters(prms),
Objective = _helperObjective.FromParameters(prms),
Learning = _helperLearning.FromParameters(prms)
};
// irrelevant parameter for managed trees which always use NaN for missing value
prms.Remove("zero_as_missing");
prms.Remove("saved_feature_importance_type");
if (prms.Count > 0)
{
Console.WriteLine($"WARNING: Unknown new parameters {String.Join(",", prms.Keys)}");
}
return (res, p);
}
