HackPrague

# Start writing code here... import pandas as pd import numpy as np from sklearn.cluster import KMeans from sklearn import preprocessing from matplotlib import pyplot as plt weekDays = ["Monday", "Tuesday", "Wednesday", "Thursday", "Friday", "Saturday", "Sunday"] numOfClusters = 4 def categories_to_columns(transaction_data, categories): for cat in categories: if (transaction_data['merchant_category'] == cat): return cat def weekdays_to_columns(transaction_data): return weekDays[pd.to_datetime(transaction_data['tx_date']).dayofweek] def normalize(user): user[['amount', 'Longitude', 'Latitude']] = preprocessing.scale(user[['amount', 'Longitude', 'Latitude']]) return user def reshape_user(user_data, categories): user_reshaped = user_data.copy() for cat in categories: user_reshaped.insert(len(user_data.columns), cat, 0) for day in weekDays: user_reshaped.insert(len(user_data.columns), day, 0) for index, row in user_data.iterrows(): cat = categories_to_columns(row, categories) user_reshaped.loc[index, cat] = 1 day = weekdays_to_columns(row) user_reshaped.loc[index, day] = 1 user_reshaped.drop('merchant_category', axis=1, inplace=True) user_reshaped.drop('tx_date', axis=1, inplace=True) return user_reshaped def get_user(data, cats, user_id): user = data[data['client_id'] == user_id].copy() return preprocess_user(user, cats) def preprocess_user(user, cats): # user.sort_values(by=["tx_date", 'amount'], ascending=[True, False], inplace=True) user.drop(['client_year_of_birth', 'client_id', 'client_gender', 'shop_type', 'shop_tags', 'merchant_category_id', 'merchant_uid', 'shop_uid', 'region', 'country', 'transaction_id'], axis=1, inplace=True) user = reshape_user(user, cats) return user def preprocess_transaction(transaction, cats): transaction.drop(['client_year_of_birth', 'client_id', 'client_gender', 'shop_type', 'shop_tags', 'merchant_category_id', 'merchant_uid', 'shop_uid', 'region', 'country', 'transaction_id'], inplace=True) transaction = reshape_user(transaction, cats) return transaction def get_distance_from_means(means, point): values = [] for mean in means: values.append(np.linalg.norm(mean - point)) return [np.linalg.norm(means - point), np.array(values).T] def cluster_user_transaction(user): kmeans = KMeans(n_clusters=numOfClusters).fit(user) centroids = kmeans.cluster_centers_ # print(kmeans.labels_) # Nice Pythonic way to get the indices of the points for each corresponding cluster mydict = {i: np.where(kmeans.labels_ == i)[0] for i in range(kmeans.n_clusters)} # Transform this dictionary into list (if you need a list as result) dictlist = [] for key, value in mydict.items(): temp = [key, value] dictlist.append(temp) return centroids, dictlist def check_transaction(data, categories, transaction): user_id = transaction['client_id'].values[0] user = get_user(data, categories, user_id) transaction2 = preprocess_user(transaction.copy(), cats) # normalize user = pd.concat([user, transaction2]) user = normalize(user) transaction2 = pd.DataFrame(user.iloc[len(user) -1]).copy().transpose() user = user[:-1] means, clusters = cluster_user_transaction(user) distance_total = np.zeros([len(user)]) distances = np.zeros([len(user),numOfClusters]) for i in range(len(user)): distance_total[i], distances[i][:] = get_distance_from_means(means, user.iloc[i].values) trans_distance_total, trans_distances = get_distance_from_means(means, transaction2.values) cluster_idx = np.where(abs(trans_distance_total - distance_total) == np.amin(abs(trans_distance_total - distance_total)))[0] this_user = data[data['client_id'] == user_id] indexes = [] for v in clusters: indexes.append(this_user.transaction_id.values[v[1]]) this_user_cluster = this_user[this_user.index.isin(indexes[cluster_idx[0]])] # print('Contains shop id: ', transaction['shop_uid'].values[0] in this_user_cluster['shop_uid'].values) return not transaction['shop_uid'].values[0] in this_user_cluster['shop_uid'].values def get_client_list(): return pd.read_csv("final_clusters.csv", index_col="client_id").index def cluster_friends(client_id): clusters = pd.read_csv("final_clusters.csv", index_col="client_id") return clusters[clusters["cluster_id"] == clusters["cluster_id"][client_id]]

data = pd.read_csv('dataset_reduced_gps.csv', sep=";") data.dropna(inplace=True) data.drop(data[data['country'] != 'CZ'].index, inplace=True) data.drop(['Unnamed: 0', 'Unnamed: 0.1'], axis=1, inplace=True) cats = data["merchant_category"].unique() idx = 25

transaction = pd.DataFrame(data.iloc[idx]).copy().transpose() transaction['Latitude'] += 1/1000000 transaction['Longitude'] += 1/1000000 transaction['amount'] *= 1.04 # transaction['shop_uid'] = ['New restaurant ID'] print("Should we recommend this transaction? ", check_transaction(data, cats, transaction))

transaction = pd.DataFrame(data.iloc[idx]).copy().transpose() transaction['Latitude'] += 1/1000000 transaction['Longitude'] += 1/1000000 transaction['amount'] *= 1.04 transaction['shop_uid'] = ['New restaurant ID'] recommend = check_transaction(data, cats, transaction) print("Should we recommend this transaction? ", recommend) if recommend: recommendations = cluster_friends(*transaction["client_id"]) print(f"We have found {len(recommendations)} users to recommend this shop") print(f"Some of them are: {list(recommendations[:5].index)}")