public void TestXGBoostEvaluateTiming()
{
var filename = @"../../../../datasets/xgboost/model_xbg_trees.txt";
var treesString = File.ReadAllText(filename);
var model = XGBoost.Create(treesString, prepareShrink: true);
var filename2 = @"../../../../datasets/xgboost/xgboost_test_cases_no_feature_names.txt";
var samplesString = File.ReadLines(filename2);
var samples = new Dictionary <string, float[]>();
foreach (var line in samplesString.Skip(1))
{
var parts = line.Split(',');
var sample = parts[0];
var featureIndex = int.Parse(parts[1]);
var value = float.Parse(parts[2]);
if (!samples.ContainsKey(sample))
{
samples.Add(sample, new float[1000]);
}
samples[sample][featureIndex] = value;
}
DoXGBoostEvaluateTimingFlatShrink(model, samples);
}
public void TestXGBoostCreate()
{
var filename = @"../../../../datasets/xgboost/model_xbg_trees.txt";
var treesString = File.ReadAllText(filename);
var model = XGBoost.Create(treesString);
// TODO: ensure correctness of each tree
}
private static XGBoost GetModel(string dataPath, short[] reorderMapping)
{
var filename = Path.Combine(dataPath, @"model_xbg_trees.txt");
var treesString = File.ReadAllText(filename);
var model = XGBoost.Create(treesString);
if (reorderMapping != null)
{
model = FeatureReorderer.ReorderXGBoost(model, reorderMapping);
}
return(model);
}
private void DoXGBoostEvaluateTimingFlatShrink(XGBoost model, Dictionary <string, float[]> samples)
{
Random r = new Random(20190524);
var toRun = samples.Select(m => m.Value)
.Concat(samples.Select(m => m.Value))
.OrderBy(m => r.Next())
.ToArray();
var userFeatures = toRun[0].Take(639).ToArray();
var jobFeatures = toRun.Select(m => m.Skip(639).ToArray()).ToArray();
var results = new float[toRun.Length];
var timer = Stopwatch.StartNew();
model.EvaluateProbabilityFlatShrink(userFeatures, jobFeatures, results);
timer.Stop();
output.WriteLine($"Time taken for {toRun.Length} evaluations: {timer.Elapsed.TotalMilliseconds} ms");
}
private void DoXGBoostEvaluateTimingFlat(XGBoost model, Dictionary <string, float[]> samples)
{
Random r = new Random(20190524);
var toRun = samples.Select(m => m.Value)
.Concat(samples.Select(m => m.Value))
.OrderBy(m => r.Next())
.ToArray();
var results = new double[toRun.Length];
var timer = Stopwatch.StartNew();
for (int i = 0; i < toRun.Length; i++)
{
results[i] = model.EvaluateProbabilityFlat(toRun[i]);
}
timer.Stop();
output.WriteLine($"Time taken for {toRun.Length} evaluations: {timer.Elapsed.TotalMilliseconds} ms");
}
private static void ModelSummary(XGBoost model)
{
var allPaths = model.Trees.SelectMany(FeatureReorderer.GetAllPaths).ToArray();
Console.WriteLine($"total paths: {allPaths.Length}");
Console.WriteLine($"average path length: {allPaths.Average(m => m.Length)}");
var pageCounts = allPaths.Select(FeatureReorderer.NumMemoryPages).ToArray();
var pages = pageCounts.GroupBy(m => m)
.Select(m => new { m.Key, Count = m.Count() })
.OrderBy(m => m.Key)
.ToArray();
foreach (var pageCount in pages)
{
Console.WriteLine($" {pageCount.Key}: {pageCount.Count}");
}
Console.WriteLine($"average page count: {pageCounts.Average(m => m)}");
}
public void TestXGBoostEvaluate()
{
var filename = @"../../../../datasets/xgboost/model_xbg_trees.txt";
var treesString = File.ReadAllText(filename);
var model = XGBoost.Create(treesString, prepareShrink: true);
var filename2 = @"../../../../datasets/xgboost/xgboost_test_cases_no_feature_names.txt";
var samplesString = File.ReadLines(filename2);
var samples = new Dictionary <string, float[]>();
foreach (var line in samplesString.Skip(1))
{
var parts = line.Split(',');
var sample = parts[0];
var featureIndex = int.Parse(parts[1]);
var value = float.Parse(parts[2]);
if (!samples.ContainsKey(sample))
{
samples.Add(sample, new float[1000]);
}
samples[sample][featureIndex] = value;
}
var timer = Stopwatch.StartNew();
const int probablity_feature_index = 840;
int i = 0;
for (; i < 2; i++)
{
foreach (var sample in samples)
{
var userFeatures = sample.Value.Take(639).ToArray();
var jobFeatures = sample.Value.Skip(639).ToArray();
var actual = model.EvaluateProbabilityFlatShrink(userFeatures, jobFeatures);
var expected = sample.Value[probablity_feature_index];
Assert.InRange(actual, expected - 1e-06, expected + 1e-06);
}
}
timer.Stop();
output.WriteLine($"Time taken for {i*samples.Count} evaluations: {timer.Elapsed.TotalMilliseconds} ms");
}
public void TestXGBoostEvaluateTimingReordered()
{
var filename1 = @"../../../../datasets/xgboost/reorder.csv";
var reorderMapping = File.ReadAllLines(filename1)
.Select(m => short.Parse(m))
.ToArray();
var filename = @"../../../../datasets/xgboost/model_xbg_trees.txt";
var treesString = File.ReadAllText(filename);
var model = XGBoost.Create(treesString);
model = FeatureReorderer.ReorderXGBoost(model, reorderMapping);
var filename2 = @"../../../../datasets/xgboost/xgboost_test_cases_no_feature_names.txt";
var samplesString = File.ReadLines(filename2);
var samples = new Dictionary <string, float[]>();
foreach (var line in samplesString.Skip(1))
{
var parts = line.Split(',');
var sample = parts[0];
var featureIndex = int.Parse(parts[1]);
if (featureIndex >= 0 && featureIndex < reorderMapping.Length)
{
featureIndex = reorderMapping[featureIndex];
}
var value = float.Parse(parts[2]);
if (!samples.ContainsKey(sample))
{
samples.Add(sample, new float[1000]);
}
samples[sample][featureIndex] = value;
}
DoXGBoostEvaluateTimingFlat(model, samples);
}