public static dynamic NewConverter()
{
var converter = new PyConverter();
using (Py.GIL())
{
//XIncref is needed, or keep the PyObject
converter.Converters.Add(PythonEngine.Eval("int").Handle, (args) => { return(args.AsManagedObject(typeof(int))); });
converter.Converters.Add(PythonEngine.Eval("str").Handle, (args) => { return(args.AsManagedObject(typeof(string))); });
converter.Converters.Add(PythonEngine.Eval("float").Handle, (args) => { return(args.AsManagedObject(typeof(double))); });
converter.Converters.Add(PythonEngine.Eval("bool").Handle, (args) => { return(args.AsManagedObject(typeof(bool))); });
converter.Converters.Add(PythonEngine.Eval("list").Handle, (args) =>
{
List <object> list = new List <object>();
for (int i = 0; i < args.Length(); i++)
{
dynamic newVal = converter.Convert(args[i]);
list.Add(newVal);
}
return(list);
});
converter.Converters.Add(PythonEngine.Eval("dict").Handle, (args) =>
{
Dictionary <string, object> dict = new Dictionary <string, object>();
var dictionaryItems = args.InvokeMethod("items");
Console.WriteLine("dictionaryItems");
foreach (dynamic keyValue in dictionaryItems)
{
dynamic newKey = converter.Convert(keyValue[0]);
dynamic newValue = converter.Convert(keyValue[1]);
dict.Add(newKey.ToString(), newValue);
}
return(dict);
});
}
return(converter);
}
/// <summary>
/// Key-Value : ProductId - Prediction Value (rate)
/// </summary>
/// <returns></returns>
public Dictionary <int, double> GetPredictions()
{
try
{
using (Py.GIL())
{
dynamic ev = getFeatures.Event2vec();
logger.Debug("Event vector:" + ev?.ToString());
Console.WriteLine("Event vector:" + ev);
//simulate a time now
int epoch1 = (int)(DateTime.Now - new DateTime(1970, 1, 1)).TotalSeconds + 1200;
Console.WriteLine("epoch:" + epoch1);
dynamic timenow1 = pythonModule.testtime(epoch1);
logger.Debug("Timenow1:" + timenow1?.ToString());
Console.WriteLine("Timenow1:" + timenow1);
dynamic td = getFeatures.get_timedelta();
logger.Debug(" Time delta" + td?.ToString());
Console.WriteLine("Time delta" + td);
// calling and showing fit parameters of the algorithm
dynamic W1 = pythonModule.W1;
logger.Debug("W1: " + W1?.ToString());
dynamic W2 = pythonModule.W2;
logger.Debug("W2: " + W2?.ToString());
dynamic W3 = pythonModule.W3;
logger.Debug("W3: " + W3?.ToString());
// importing the fit matrix ' item_context_pred" founded under python code
dynamic ict = pythonModule.item_context_pred;
logger.Debug("predicted context: " + ict?.ToString());
// importing the fit matrix ' Products" founded under python code
dynamic products = pythonModule.Products;
logger.Debug("Products: " + ict?.ToString());
// checking python feature engineering functions before fitting inputs to the model
// input derived from contexts
dynamic ctoi = pythonModule.Context2Item(ict);
logger.Debug("Sale probability for each item according to the actual context: " + ctoi?.ToString());
// input derived from user profile
dynamic uti = pythonModule.User2Item(this.userProfile);
logger.Debug("Sale probability for each item according to the user_profile: " + uti?.ToString());
// input derived from happening events
dynamic eti = pythonModule.Event2Item(ev);
logger.Debug("Sale probability for each item according to happening event: " + eti?.ToString());
Console.WriteLine("Sale probability for each item according to happening event: " + eti?.ToString());
// input derived from time delta
dynamic mah = pythonModule.Mah(td);
logger.Debug("Sale probability for each item according to the visit duration of the client: " + mah?.ToString());
Console.WriteLine("Sale probability for each item according to the visit duration of the client: " + mah?.ToString());
// putting all together and using the python class "Recommendation" to make predictions
dynamic rec = pythonModule.Recommendation(ict, products, userProfile, ev, td, W1, W2, W3);
dynamic pred;
if (UsedPredictionMode == PredictionMode.PredictionWithProductScoreOptimization)
{
logger.Debug("SCORE-BASED PREDICTION:");
pred = rec.get_prediction(out_format: "dict");// out_format: " dict" gives the output under a dictionary
Console.WriteLine("SCORE-BASED PREDICTION:");
Console.WriteLine(pred);
//dynamic predtodict = pred.to_dict();
}
else
{
logger.Debug("PRICE-BASED PREDICTION:");
pred = rec.get_optprediction();// out_format: " dict" gives the uoutput under a dictionary
Console.WriteLine("PRICE-BASED PREDICTION:");
Console.WriteLine(pred);
//dynamic predtodict = pred.to_dict();
}
if (pred != null)
{
logger.Debug(pred?.ToString());
}
var converter = PyConverter.NewConverter(); //create an instance of PyConverter, take from ConverteurPy.cs script
Dictionary <string, object> productNameAndScores = converter.Convert(pred); //b is a List of CLR objects
//
// Transform product Name into equivalent product Id:
//
Dictionary <int, double> productIdAndScores = new Dictionary <int, double>();
if (productNameAndScores != null)
{
using (ShopDBModel context = new ShopDBModel())
{
//List<ProductScore> productScores = productNameAndScores.Select(pns => new ProductScore { Name = pns.Key, Score = pns.Value, Id = 0 }).ToList();
//List<string> productNames = productScores.Select(ps => ps.Name).ToList();
//// Retrieve corresponding products from database:
//List<ProductScore> productsFromDB = context.Products.Where(p => productNames.Contains(p.Name)).Select(p2 => new ProductScore { Id = p2.ID, Name = p2.Name }).ToList();
//ProductScore currentProductScore = null;
//// Change product Name with product ID:
//foreach (KeyValuePair<string, double> kvp in productNameAndScores)
//{
// currentProductScore = productsFromDB.FirstOrDefault(pfd => pfd.Name == kvp.Key);
// if (currentProductScore == null)
// {
// continue;
// }
// productIdAndScores.Add(currentProductScore.Id, kvp.Value);
//}
}
}
return(productIdAndScores);
}
}
catch (Exception ex)
{
logger.Error(nameof(GetPredictions), "Unable to compute user predictions", ex);
return(null);
}
}
