def setCFtype(self, cftype):
self.CFtype = cftype
foldername, filename = "Models", self.CFtype + "_CFModel.pkl"
if (self.fitModeAtInitialize == True):
if (AlgoBase.checkPathExists(self, foldername, filename) == True):
self.preds_df = AlgoBase.loadModel(self, foldername, filename)
else:
self.fit(similarity="cosine", CFtype="user")
self.fit(similarity="cosine", CFtype="item")
def fit(self, similarity="cosine"):
try:
corpus = self.BooksM[self.BooksM.ISBN.isin(self.RatingsM.ISBN)]["Book-Title"].tolist()
vectorizer = TfidfVectorizer()
X = vectorizer.fit_transform(corpus)
self.X_df = pd.DataFrame(X.todense(), columns = vectorizer.get_feature_names(), index=corpus)
self.Similarity_Matrix = self.Compute_Similarity(similarity)
AlgoBase.saveModel(self, self.Similarity_Matrix, "Models", "TfidfModel.pkl")
except Exception as e:
print(e)
def __init__(self, dsReader, fitModeAtInitialize=True):
AlgoBase.__init__(self, dsReader, fitModeAtInitialize)
self.RECOMMENDER_NAME = "SVD Recommender"
foldername, filename = "Models", "SVDModel.pkl"
if (self.fitModeAtInitialize == True):
if (AlgoBase.checkPathExists(self, foldername, filename) == True):
self.preds_df = AlgoBase.loadModel(self, foldername, filename)
else:
self.fit()
def __init__(self, dsReader, fitModeAtInitialize=True):
AlgoBase.__init__(self, dsReader, fitModeAtInitialize)
self.RECOMMENDER_NAME = "TF=IDF Recommender"
foldername, filename = "Models", "TfidfModel.pkl"
if(self.fitModeAtInitialize == True):
if(AlgoBase.checkPathExists(self, foldername, filename) == True):
self.similarityM = AlgoBase.loadModel(self, foldername, filename)
else:
self.fit()
def fit(self, similarity="cosine", CFtype="user"):
self.similarity = similarity
self.CFtype = CFtype
self.Similarity_Matrix = self.Compute_Similarity(
self.similarity, self.train, self.CFtype)
self.user_prediction = self.predict(self.train, self.Similarity_Matrix,
self.CFtype)
self.preds_df = pd.DataFrame(self.user_prediction,
columns=self.URM.columns,
index=self.URM.index)
AlgoBase.saveModel(self, self.preds_df, "Models",
self.CFtype + "_CFModel.pkl")
def __init__(self, dsReader, fitModeAtInitialize=True):
AlgoBase.__init__(self, dsReader, fitModeAtInitialize)
self.RECOMMENDER_NAME = "Embeddings Recommender"
# Define N and P - number of users and books respectively
self.N = len(self.RatingsM["User-ID"].unique().tolist())
self.P = len(self.RatingsM["ISBN"].unique().tolist())
foldername, filename = "Models", "EMBModel.h5"
self.filepath = os.path.dirname(os.path.abspath(__file__))
self.filepath = self.filepath + "/" + foldername + "/" + filename
if (self.fitModeAtInitialize == True):
if (AlgoBase.checkPathExists(self, foldername, filename) == True):
#self.model = AlgoBase.loadModel(self, foldername, filename)
self.model = tf.keras.models.load_model(self.filepath)
else:
self.fit()
def fit(self):
U, s, Vt = svds(self.train, k=100)
sigma = np.diag(s)
# Check number of terms of s to include to capture at least 90% of the information
sum = 0
tot = s.sum()
ctr = 0
while (sum < 0.9):
sum += s[ctr] / tot
ctr += 1
print("K: {}, Cumulative Variance: {}".format(ctr, sum))
self.all_user_predicted_ratings = np.dot(np.dot(U, sigma), Vt)
self.preds_df = pd.DataFrame(self.all_user_predicted_ratings,
columns=self.URM.columns,
index=self.URM.index)
AlgoBase.saveModel(self, self.preds_df, "Models", "SVDModel.pkl")
def compareActualPredictedRatings(self, user_id, userRecBookList):
userRecBookList_ISBNs = list(userRecBookList.keys())
# Get and sort the user's predictions
sorted_user_predictions = self.preds_df.loc[user_id].sort_values(
ascending=False)
predictedRatings = pd.DataFrame(sorted_user_predictions).reset_index()
# Get predicted ratings for user recommended booklist
predictedRatings = predictedRatings[predictedRatings["ISBN"].isin(
userRecBookList_ISBNs)].rename(
columns={user_id: 'Predicted-Rating'})
# Get actual ratings for user recommended booklist
actualRatings = self.dsReader.getRatingsByISBNs(
user_id, userRecBookList_ISBNs)
# Merge both predicted and actual ratings
allRatings = actualRatings.merge(predictedRatings,
how='left',
left_on='ISBN',
right_on='ISBN')
# Get book details based on ISBN
allRatings = allRatings.merge(
self.BooksM[['ISBN', 'Book-Title', 'Book-Author']],
how='left',
left_on='ISBN',
right_on='ISBN')
allRatings = allRatings[[
'ISBN', 'Book-Title', 'Book-Author', 'Predicted-Rating',
'Book-Rating'
]]
# Evaluate RMSE metric for predicted/actual ratings
predicted = np.asarray(allRatings["Predicted-Rating"].tolist())
actual = np.asarray(allRatings["Book-Rating"].tolist())
eval_res = AlgoBase.Evaluate(self, predicted, actual, ["rmse"])
return allRatings, eval_res
def __init__(self, dsReader, fitModeAtInitialize=True):
AlgoBase.__init__(self, dsReader, fitModeAtInitialize)
self.RECOMMENDER_NAME = "User-Item Based Collaborative Filtering Recommender"