Code fragments of main.py

• # See for reference: https://naomy-gomes.medium.com/the-cosine-similarity-and-its-use-in-recommendation-systems-cb2ebd811ce1
• import pandas as pd
• from helper import preprocess, cosine_sim, get_combinations, show_welcome_text
• import psycopg2
• import json
• import os
• ''' This is a simple supplier recommendation system using cosine similarity to provide recommendations for a specific rating input based on the the most similar rating entries. '''
• # get ratings from database
• conn = psycopg2.connect(database=os.environ["DB_NAME"], user=os.environ["DATA_SOURCE_USERNAME"], host=os.environ["DATA_SOURCE_HOST"], password=os.environ["DATA_SOURCE_PASSWORD"], port=os.environ["DATA_SOURCE_PORT"] )
• conn = psycopg2.connect(database=os.environ["DB_NAME"], user=os.environ["DATA_SOURCE_USERNAME"], host=os.environ["DATA_SOURCE_HOST"], password=os.environ["DATA_SOURCE_PASSWORD"], port=os.environ["DATA_SOURCE_PORT"] )
• cursor = conn.cursor()
• cursor.execute("SELECT * FROM supplier_data")
• ratings = cursor.fetchall()
• column_names = [desc[0] for desc in cursor.description]
• df = pd.DataFrame(ratings, columns=column_names)
• def recommend(input_order_index, df):
• """ Content based recommendations Returns the most similar suppliers """
• # create vector and temp df without id columns
• df_temp = df.copy()
• cols = ['rating_id'] + df_temp.columns[df_temp.iloc[-1].isna()].tolist()
• # gets corresponding values of rating_id
• value_vec = df.loc[input_order_index]
• # drop specified columns
• value_vec = value_vec.drop(index = cols)
• value_vec = value_vec.values
• # gets corresponding values of rating_id
• df_temp.drop(columns = cols, inplace = True)
• # drop last column
• df_temp.drop(df_temp.tail(1).index, inplace=True)
• df.drop(df.tail(1).index, inplace=True)
• # calculate similarity of input rating_id vector and all other vectors
• # create a column containing the values of the cosine similarity
• df_temp['sim']= df_temp.apply(lambda x: cosine_sim(value_vec,x.values), axis=1)
• # goes through all the rating vectors in the table
• df = pd.merge(df_temp, df, how='inner',on=None, validate=None)
• # merge temp df with similarity column into df
• # drop the last row
• # gets only the X most similar rows
• df_rec = df.nlargest(4, columns='sim')
• df_final = df_rec.sort_values("sim", ascending=False)
• # order by highest similarity value
• print("\nFound Similarities:\n", df_final)
• return df_final
• def calc_total_rating(df, supplier):
• filtered_df = df[df['supplier_name'] == supplier]
• in_time_count = filtered_df[filtered_df['rating'] == 'in time'].shape[0]
• total_count = round((in_time_count/filtered_df.shape[0])*100, 2)
• return total_count
• def add_ratings(df, df_original):
• supplier_dict = df.to_dict()
• supplier_list = [{key: s, "ratings": calc_total_rating(df_original, s)} for (key, s) in supplier_dict.items()]
• return supplier_list
• def supplier_recommend(input_order, df): # df = pd.read_json(r'testratings.json')
• df_original = df.copy()
• input_order_df = pd.DataFrame([input_order], columns=df.columns)
• # append the new row to df
• df = pd.concat([df, input_order_df], ignore_index=True)
• input_order_index = df.shape[0] - 1
• # add index column
• df['rating_id'] = df.index
• df = preprocess(df)
• # call recommendation fct
• df_recommendations = recommend(input_order_index, df)
• # search supplier name for the recommended IDs, drop duplicates
• df_suppliers = pd.merge(df_recommendations, df_original, how='inner',on=None, validate=None)["supplier_name"]
• df_suppliers.drop_duplicates(inplace=True)
• list_suppliers = add_ratings(df_suppliers, df_original)
• return list_suppliers
• show_welcome_text()
• combinations = get_combinations()
• recommended_supplier_list = []
• for combi in combinations:
• list_suppliers = supplier_recommend(combi, df)
• recommended_supplier_list.append([combi, list_suppliers])
• print(recommended_supplier_list)
• result_file = 'result.json'
• with open(result_file, 'w') as f:
• json.dump(recommended_supplier_list, f)

Graph