// Compare Recommendations (Naive Bayes vs. Modified KNN)
static void CompareRecommendations()
{
UserRequest request = new UserRequest();
// input recipe examples
string[] first = new string[2] {
"apple", "sugar"
};
string[] second = new string[4] {
"bacon", "chicken", "salt", "black pepper"
};
string[] third = new string[4] {
"soy sauce", "ginger", "chicken", "water"
};
string[] fourth = new string[4] {
"mushrooms", "chicken breast", "chicken broth", "sugar"
};
string[] fifth = new string[7] {
"confectioners sugar", "cocoa powder", "condensed milk", "applesauce", "almond extract", "cheddar cheese", "buttermilk"
};
ModelChoice[] modelChoices = new ModelChoice[2] {
ModelChoice.NB, ModelChoice.MKNN
};
foreach (ModelChoice item in modelChoices)
{
// recommend ingredients to ADD
request.TopRecommendations(10, first, item, true, false);
request.TopRecommendations(10, second, item, true, false);
request.TopRecommendations(10, third, item, true, false);
// recommend ingredients to REMOVE
request.TopRecommendations(10, fourth, item, false, true);
request.TopRecommendations(10, fifth, item, false, true);
}
}
// Get Recipes
public Data[] GetRecipes(ModelChoice model_choice, DataPurpose data_purpose)
{
MLContext mlContext = new MLContext();
IDataView testData = GetDataView(model_choice, mlContext, data_purpose);
IDataView features = GetDataView(model_choice, mlContext, DataPurpose.FEATURES);
return(GetData(testData, features));
}
// Evaluate different models
static void Evaluate(ModelChoice choice)
{
Evaluation eval = new Evaluation();
// Evaluate Non-negative Matrix Factorization (12 minutes)
if (choice.Equals(ModelChoice.NMF))
{
Recommender nmf = new Recommender();
eval.EvaluateNMF(nmf.GetModel());
}
// Evaluate Naive Bayes (5 seconds each)
else if (choice.Equals(ModelChoice.NB))
{
NaiveBayes nb = new NaiveBayes();
double[][] model = nb.GetModel();
Console.WriteLine("1: Evaluate ALL versions of Naive Bayes");
Console.WriteLine("2: Evaluate only the BEST version of Naive Bayes (normalization, laplace, uniform prior)");
string option = Console.ReadLine();
// ALL
if (option == "1")
{
eval.EvaluateNB(model, false, false, true);
eval.EvaluateNB(model, true, false, true); // with laplace smoothing
eval.EvaluateNB(model, false, true, true); // with normalization (best results)
eval.EvaluateNB(model, true, true, true); // with laplace smoothing and normalization
// uniform prior
eval.EvaluateNB(model, false, false, false);
eval.EvaluateNB(model, true, false, false); // with laplace smoothing
eval.EvaluateNB(model, false, true, false); // with normalization (best results)
}
// BEST
eval.EvaluateNB(model, true, true, false); // with laplace smoothing, normalization, and uniform prior
}
// Evaluate KNN (35 seconds)
else if (choice.Equals(ModelChoice.KNN))
{
eval.EvaluateKNN(6, DistanceChoice.Jaccard, Voting.Unweighted);
}
// Evaluate Modified KNN (35 seconds)
else
{
eval.EvaluateModifiedKNN(DistanceChoice.Jaccard_Similarity, Voting.Unweighted);
}
}
// Get ingredients sorted by what is most recommended
public Recommendation[] GetRecommendations(ModelChoice model_choice, int[] recipe)
{
Recommendation[] recommendations = null;
if (model_choice.Equals(ModelChoice.NB))
{
NaiveBayes nb = new NaiveBayes();
recommendations = nb.RecipeRecommendations(nb.GetModel(), recipe, true, true, false);
}
else if (model_choice.Equals(ModelChoice.KNN))
{
KNN knn = new KNN();
recommendations = knn.GetRecommendations(6, DistanceChoice.Jaccard, recipe, Voting.Unweighted);
}
else if (model_choice.Equals(ModelChoice.MKNN))
{
KNN knn = new KNN();
int new_k = 0;
recommendations = knn.GetRecommendations_ModifiedKNN(recipe, DistanceChoice.Jaccard_Similarity, Voting.Unweighted, ref new_k);
}
return(recommendations);
}
// Get DataView
public IDataView GetDataView(ModelChoice choice, MLContext ml, DataPurpose purpose)
{
// get features
if (purpose.Equals(DataPurpose.FEATURES))
{
return(LoadFeatures(Path.Combine(Environment.CurrentDirectory, "Data", FEATURES_DATA), ml));
}
// get NMF data (entries set aside for testing)
else if (choice.Equals(ModelChoice.NMF))
{
// test data
if (purpose.Equals(DataPurpose.TEST))
{
return(LoadData(Path.Combine(Environment.CurrentDirectory, "Data", NMF_TEST_DATA), ml));
}
// training data
else
{
return(LoadData(Path.Combine(Environment.CurrentDirectory, "Data", NMF_TRAIN_DATA), ml));
}
}
// get data (recipes set aside for testing)
else
{
// test data
if (purpose.Equals(DataPurpose.TEST))
{
return(LoadData(Path.Combine(Environment.CurrentDirectory, "Data", TEST_DATA), ml));
}
// training data
else
{
return(LoadData(Path.Combine(Environment.CurrentDirectory, "Data", TRAIN_DATA), ml));
}
}
}
// Ingredients to ADD to or REMOVE from a recipe
public void TopRecommendations(int top, string[] recipe_str, ModelChoice model_choice, bool add, bool include_recipe_ingrs)
{
DataManager dm = new DataManager();
// get training data
Data[] data = dm.GetRecipes(model_choice, DataPurpose.TRAIN);
Console.WriteLine("You model choice: " + model_choice.ToString());
// input recipe
int[] recipe = new int[recipe_str.Length];
for (int i = 0; i < recipe_str.Length; i++)
{
try
{
// trim and make lowercase
recipe_str[i] = recipe_str[i].Trim().ToLower();
// find ingredient
recipe[i] = data.Where(d => d.ingredient.name.Equals(recipe_str[i])).ToArray()[0].ingredient.id;
}
catch
{
// get features (ingredients)
string[] features = GetAllIngredients(false);
bool found = false;
// try finding a similar ingredient
foreach (string ingr in features)
{
if (ingr.StartsWith(recipe_str[i]) || ingr.Contains(recipe_str[i]))
{
recipe_str[i] = ingr;
recipe[i] = data.Where(d => d.ingredient.name.Equals(ingr)).ToArray()[0].ingredient.id;
found = true;
break;
}
}
// ingredient not found
if (found == false)
{
Console.WriteLine("Ingredient [" + recipe_str[i] + "] was not found");
return;
}
}
}
Console.Write("Your recipe: ");
PrintRecipe(recipe_str);
// keep track of ingredient recommendations
Recommendation[] recommendations = null;
// Naive Bayes
if (model_choice.Equals(ModelChoice.NB))
{
NaiveBayes nb = new NaiveBayes();
recommendations = nb.RecipeRecommendations(nb.GetModel(), recipe, true, true, false);
}
// k Nearest Neighbors
else if (model_choice.Equals(ModelChoice.KNN))
{
KNN knn = new KNN();
recommendations = knn.GetRecommendations(6, DistanceChoice.Jaccard, recipe, Voting.Unweighted);
}
// Modified k Nearest Neighbors (dynamic k)
else if (model_choice.Equals(ModelChoice.MKNN))
{
KNN knn = new KNN();
int new_k = 0;
recommendations = knn.GetRecommendations_ModifiedKNN(recipe, DistanceChoice.Jaccard_Similarity, Voting.Unweighted, ref new_k);
}
else
{
return;
}
// Ingredients to Add
if (add == true)
{
Console.WriteLine("Your recommendations:");
for (int i = 0; i < top; i++)
{
// skip ingredients in recipe
if (include_recipe_ingrs == false && recipe_str.Contains(recommendations[i].ingredient.name))
{
top++;
continue;
}
Console.WriteLine(recommendations[i].ingredient.name);
}
}
// Ingredients to Remove
else
{
Console.WriteLine("Ingredients ordered by what to remove first:");
// only keep scores of ingredients in input recipe
recommendations = recommendations.Where(d => recipe.Contains(d.ingredient.id)).ToArray();
// sort by score
recommendations = recommendations.OrderBy(d => d.score).ToArray();
for (int i = 0; i < recipe.Length; i++)
{
Console.WriteLine(recommendations[i].ingredient.name);
}
}
Console.WriteLine();
}