static void Main(string[] args)
{
Console.Write("Enter command:");
var cmd = Console.ReadLine();
switch (cmd)
{
case "flat":
FlatternUsers();
break;
case "goup_class":
ExtractGroupInfo();
break;
case "antibot":
AntiBot();
break;
case "repack":
Repack();
break;
case "reveal":
Reveal();
break;
case "quest_fifth":
QuestionaireDatasetPreparation.StoreDatasetFifth();
break;
case "train_fifth":
Fifth();
break;
case "compute_fifth":
ComputeFifth();
break;
case "clean":
Clean();
break;
case "katya":
Katya();
break;
case "peter":
Peter();
break;
case "denis":
Denis();
break;
case "task":
ComputeTask();
break;
}
}
private static Fifther Fifth()
{
var groupmapping = JsonConvert.DeserializeObject <Dictionary <long, double[]> >(File.ReadAllText("map_groups.json"));
var bagOfTerms = File.ReadAllLines("res\\expert_topics.csv").Skip(1).Select(z => z.Split(',')).ToDictionary(z => int.Parse(z[0]), z => new HashSet <string>(z[1].Split(' ')));
var xFilename = "224053984_dataset.json";
var scope = new QuestionnaireScope();
XmlSerializer serializer = new XmlSerializer(typeof(QuestionnaireScope));
using (var reader = new StreamReader("test_fifth.xml"))
{
scope = (QuestionnaireScope)serializer.Deserialize(reader);
}
var log = new LoggerConfiguration()
.MinimumLevel.Verbose()
.WriteTo.LiterateConsole()
.CreateLogger();
var repository = new UserGetRepository("passed_tests_ferrets.s3db", log, new CompressorProto());
Dictionary <int, float[]> vectors = new Dictionary <int, float[]>();
var users = new UserGet[0];
var count = 0;
const int batch = 1000;
do
{
try
{
users = repository.RangeSelect(count, batch).ToArray();
foreach (var user in users)
{
vectors.Add(user.id, user.ToVector(groupmapping, bagOfTerms));
}
count += batch;
log.Information("Done {Count} recs.", count);
}
catch (Exception ex)
{
Console.WriteLine(ex);
}
} while (users.Length != 0);
//File.WriteAllText("datax.csv", string.Join(", ", FlatUsertToVectorMapping.GetHeader(groupmapping.FirstOrDefault().Value.Length, bagOfTerms.Count)) + "\r\n");
//File.AppendAllLines("datax.csv", vectors.Select(z => string.Join(", ", z.Value.Select(x => x.ToString("0.000", CultureInfo.InvariantCulture)))));
var X = JsonConvert.DeserializeObject <List <FifthAttendance> >(File.ReadAllText(xFilename)).Where(z => vectors.ContainsKey(z.vkid)).Where(z => QuestionaireDatasetPreparation.CovertAnswersToVector(scope, z.AnswersId).Length == 120).ToArray();
var Y = X.Select(z => QuestionaireDatasetPreparation.CovertAnswersToVector(scope, z.AnswersId)).ToArray();
File.WriteAllLines("answers.csv", Y.Select(z => string.Join(", ", z)));
var rnd = new Random(Environment.TickCount);
var train = Enumerable.Range(0, X.Length).OrderBy(z => rnd.NextDouble()).ToArray();
var x_train = train.Take(X.Length * 80 / 100).Select(z => vectors[X[z].vkid]).ToArray();
var y_train = train.Take(X.Length * 80 / 100).Select(z => Y[z]).ToArray();
var x_test = train.Skip(X.Length * 80 / 100).Select(z => vectors[X[z].vkid]).ToArray();
var y_test = train.Skip(X.Length * 80 / 100).Select(z => Y[z]).ToArray();
var fifther = new Fifther();
Console.WriteLine();
for (int qnum = 0; qnum < 120; qnum++)
{
Console.WriteLine($"Question: {qnum}");
var yds = y_train.Select(z => (float)z[qnum]).ToArray();
var ytds = y_test.Select(z => (float)z[qnum]).ToArray();
var parameters = new Dictionary <string, object>();
parameters["max_depth"] = 10;
parameters["learning_rate"] = 0.1f;
parameters["n_estimators"] = 300;
parameters["silent"] = true;
parameters["objective"] = "multi:softprob";//"binary:logistic";//
parameters["nthread"] = -1;
parameters["gamma"] = 4f;
parameters["min_child_weight"] = 2;
parameters["max_delta_step"] = 1;
parameters["subsample"] = 1f;
parameters["colsample_bytree"] = 1f;
parameters["colsample_bylevel"] = 1f;
parameters["reg_alpha"] = 0f;
parameters["reg_lambda"] = 1f;
parameters["scale_pos_weight"] = 1f;
parameters["base_score"] = 0.8F;
parameters["seed"] = 0;
parameters["missing"] = float.NaN;
parameters["num_class"] = 5;
var xgbc = new XGBClassifier(parameters);
xgbc.Fit(x_train, yds);
fifther.AddLevel(qnum, xgbc);
var discrepancy = 0.0;
var dist = 0.0;
var preds = xgbc.PredictDistr(x_train);
for (int pos = 0; pos < preds.Length; pos++)
{
var tmp = new float[5];
tmp[(int)yds[pos]] = 1f;
dist += Math.Abs(det.GetMaxIndex(preds[pos]) - yds[pos]);
discrepancy += det.EuclidianDistance(tmp, preds[pos]);
}
Console.WriteLine("[Train] Discrepancy {0:0.000} Dist {1:0.000}", 1.0 - discrepancy / (preds.Length * Math.Sqrt(2.0)), dist / preds.Length);
preds = xgbc.PredictDistr(x_test);
discrepancy = 0.0;
dist = 0.0;
for (int pos = 0; pos < preds.Length; pos++)
{
var tmp = new float[5];
tmp[(int)yds[pos]] = 1f;
dist += Math.Abs(det.GetMaxIndex(preds[pos]) - yds[pos]);
discrepancy += det.EuclidianDistance(tmp, preds[pos]);
}
Console.WriteLine("[Test ] Discrepancy {0:0.000} Dist {1:0.000}", 1.0 - discrepancy / (preds.Length * Math.Sqrt(2.0)), dist / preds.Length);
}
Console.WriteLine("Done");
fifther.Save(Path.Combine(AppDomain.CurrentDomain.BaseDirectory, "fifth"));
Console.ReadLine();
Console.ReadLine();
Console.ReadLine();
Console.ReadLine();
return(fifther);
}
private static void ComputeFifth()
{
var fifther = new Fifther();
fifther.Load(Path.Combine(AppDomain.CurrentDomain.BaseDirectory, "fifth"));
Console.Write("Enter dataset:");
var cString = Console.ReadLine();
var bagOfTerms = File.ReadAllLines("res\\expert_topics.csv").Skip(1).Select(z => z.Split(',')).ToDictionary(z => int.Parse(z[0]), z => new HashSet <string>(z[1].Split(' ')));
var scope = new QuestionnaireScope();
XmlSerializer serializer = new XmlSerializer(typeof(QuestionnaireScope));
using (var reader = new StreamReader("test_fifth.xml"))
{
scope = (QuestionnaireScope)serializer.Deserialize(reader);
}
var log = new LoggerConfiguration()
.MinimumLevel.Verbose()
.WriteTo.LiterateConsole()
.CreateLogger();
var repository = new UserGetRepository(cString, log, new CompressorProto());
var fifthRepo = new FifthResultRepository(cString);
var groupmapping = JsonConvert.DeserializeObject <Dictionary <long, double[]> >(File.ReadAllText("map_groups.json"));
fifthRepo.CleanAll();
var count = 0;
var gsw = Stopwatch.StartNew();
UserGet[] users = new UserGet[0];
const int batch = 1000;
do
{
try
{
users = repository.RangeSelect(count, batch).ToArray();
users = users.Where(z => z?.Groups?.Count > 0).ToArray();
float[][] input = new float[users.Length][];
for (int user = 0; user < users.Length; user++)
{
input[user] = users[user].ToVector(groupmapping, bagOfTerms);
}
var preds = fifther.PredictDistr(input, 5);
fifthRepo.Insert(users.Select(z => z.id).ToArray(), preds.Select(z => QuestionaireDatasetPreparation.PredictionsToScales(scope, z)).ToArray());
count += batch;
log.Information("Done {Count} recs. {DaysForMillion} days", count, TimeSpan.FromMilliseconds(1000000 * gsw.ElapsedMilliseconds / count).TotalDays);
}
catch (Exception ex)
{
Console.WriteLine(ex);
}
} while (users.Length != 0);
Console.WriteLine("Done");
Console.ReadLine();
Console.ReadLine();
}