public static void Main(string[] args)
{
AppDomain.CurrentDomain.UnhandledException += new UnhandledExceptionEventHandler(Handlers.UnhandledExceptionHandler);
// check number of command line parameters
if (args.Length < 4)
Usage("Not enough arguments.");
// read command line parameters
RecommenderParameters parameters = null;
try { parameters = new RecommenderParameters(args, 4); }
catch (ArgumentException e) { Usage(e.Message); }
// other parameters
string data_dir = parameters.GetRemoveString( "data_dir");
//Console.Error.WriteLine("data_dir " + data_dir);
string item_attributes_file = parameters.GetRemoveString( "item_attributes");
string user_attributes_file = parameters.GetRemoveString( "user_attributes");
//string save_mapping_file = parameters.GetRemoveString( "save_model");
int random_seed = parameters.GetRemoveInt32( "random_seed", -1);
bool no_eval = parameters.GetRemoveBool( "no_eval", false);
bool compute_fit = parameters.GetRemoveBool( "compute_fit", false);
string prediction_file = parameters.GetRemoveString( "prediction_file");
if (random_seed != -1)
MyMediaLite.Util.Random.InitInstance(random_seed);
// main data files and method
string trainfile = args[0].Equals("-") ? "-" : Path.Combine(data_dir, args[0]);
string testfile = args[1].Equals("-") ? "-" : Path.Combine(data_dir, args[1]);
string load_model_file = args[2];
string method = args[3];
// set correct recommender
switch (method)
{
case "MF-ItemMapping":
recommender = Recommender.Configure(mf_map, parameters, Usage);
break;
// case "MF-ItemMapping-Optimal":
// recommender = Recommender.Configure(mf_map_opt, parameters, Usage);
// break;
// case "BPR-MF-ItemMapping-kNN":
// recommender = Recommender.Configure(mf_map_knn, parameters, Usage);
// break;
// case "BPR-MF-ItemMapping-SVR":
// recommender = Recommender.Configure(mf_map_svr, parameters, Usage);
// break;
default:
Usage(string.Format("Unknown method: '{0}'", method));
break;
}
if (parameters.CheckForLeftovers())
Usage(-1);
// TODO move loading into its own method
// ID mapping objects
EntityMapping user_mapping = new EntityMapping();
EntityMapping item_mapping = new EntityMapping();
// training data
training_data = MyMediaLite.IO.RatingPrediction.Read(Path.Combine(data_dir, trainfile), user_mapping, item_mapping);
recommender.Ratings = training_data;
// user attributes
if (recommender is IUserAttributeAwareRecommender)
{
if (user_attributes_file.Equals(string.Empty))
Usage("Recommender expects user_attributes=FILE.");
else
((IUserAttributeAwareRecommender)recommender).UserAttributes = AttributeData.Read(Path.Combine(data_dir, user_attributes_file), user_mapping);
}
// item attributes
if (recommender is IItemAttributeAwareRecommender)
{
if (item_attributes_file.Equals(string.Empty))
Usage("Recommender expects item_attributes=FILE.");
else
((IItemAttributeAwareRecommender)recommender).ItemAttributes = AttributeData.Read(Path.Combine(data_dir, item_attributes_file), item_mapping);
}
// test data
test_data = MyMediaLite.IO.RatingPrediction.Read( Path.Combine(data_dir, testfile), user_mapping, item_mapping );
TimeSpan seconds;
Recommender.LoadModel(recommender, load_model_file);
// set the maximum user and item IDs in the recommender - this is important for the cold start use case
recommender.MaxUserID = user_mapping.InternalIDs.Max();
recommender.MaxItemID = item_mapping.InternalIDs.Max();
Console.Error.WriteLine(string.Format(CultureInfo.InvariantCulture, "ratings range: [{0}, {1}]", recommender.MinRating, recommender.MaxRating));
DisplayDataStats();
Console.Write(recommender.ToString() + " ");
if (compute_fit)
{
seconds = Utils.MeasureTime( delegate() {
int num_iter = recommender.NumIterMapping;
recommender.NumIterMapping = 0;
recommender.LearnAttributeToFactorMapping();
Console.Error.WriteLine();
Console.Error.WriteLine(string.Format(CultureInfo.InvariantCulture, "iteration {0} fit {1}", -1, recommender.ComputeFit()));
recommender.NumIterMapping = 1;
for (int i = 0; i < num_iter; i++, i++)
{
recommender.IterateMapping();
Console.Error.WriteLine(string.Format(CultureInfo.InvariantCulture, "iteration {0} fit {1}", i, recommender.ComputeFit()));
}
recommender.NumIterMapping = num_iter; // restore
} );
}
else
{
seconds = Utils.MeasureTime( delegate() {
recommender.LearnAttributeToFactorMapping();
} );
}
Console.Write("mapping_time " + seconds + " ");
if (!no_eval)
seconds = EvaluateRecommender(recommender);
Console.WriteLine();
if (prediction_file != string.Empty)
{
Console.WriteLine();
seconds = Utils.MeasureTime(
delegate() {
Prediction.WritePredictions(recommender, test_data, user_mapping, item_mapping, prediction_file);
}
);
Console.Error.WriteLine("predicting_time " + seconds);
}
}
static TimeSpan EvaluateRecommender(MF_Mapping recommender)
{
Console.Error.WriteLine(string.Format(CultureInfo.InvariantCulture, "fit {0}", recommender.ComputeFit()));
TimeSpan seconds = Utils.MeasureTime( delegate()
{
var result = MyMediaLite.Eval.Ratings.Evaluate(recommender, test_data);
MyMediaLite.Eval.Ratings.DisplayResults(result);
} );
Console.Write(" testing " + seconds);
return seconds;
}
