workshops_recommender_systems/P3. k-nearest neighbours.ipynb

Self made simplified I-KNN

import helpers
import pandas as pd
import numpy as np
import scipy.sparse as sparse
from collections import defaultdict
from itertools import chain
import random

train_read=pd.read_csv('./Datasets/ml-100k/train.csv', sep='\t', header=None)
test_read=pd.read_csv('./Datasets/ml-100k/test.csv', sep='\t', header=None)
train_ui, test_ui, user_code_id, user_id_code, item_code_id, item_id_code = helpers.data_to_csr(train_read, test_read)

class IKNN():
    
    def fit(self, train_ui):
        self.train_ui=train_ui
        
        train_iu=train_ui.transpose()
        norms=np.linalg.norm(train_iu.A, axis=1) # here we compute lenth of each item ratings vector
        norms=np.vectorize(lambda x: max(x,1))(norms[:,None]) # to avoid dividing by zero

        normalized_train_iu=sparse.csr_matrix(train_iu/norms)

        self.similarity_matrix_ii=normalized_train_iu*normalized_train_iu.transpose()
        
        self.estimations=np.array(train_ui*self.similarity_matrix_ii/((train_ui>0)*self.similarity_matrix_ii))
        
    def recommend(self, user_code_id, item_code_id, topK=10):
        
        top_k = defaultdict(list)
        for nb_user, user in enumerate(self.estimations):
            
            user_rated=self.train_ui.indices[self.train_ui.indptr[nb_user]:self.train_ui.indptr[nb_user+1]]
            for item, score in enumerate(user):
                if item not in user_rated and not np.isnan(score):
                    top_k[user_code_id[nb_user]].append((item_code_id[item], score))
        result=[]
        # Let's choose k best items in the format: (user, item1, score1, item2, score2, ...)
        for uid, item_scores in top_k.items():
            item_scores.sort(key=lambda x: x[1], reverse=True)
            result.append([uid]+list(chain(*item_scores[:topK])))
        return result
    
    def estimate(self, user_code_id, item_code_id, test_ui):
        result=[]
        for user, item in zip(*test_ui.nonzero()):
            result.append([user_code_id[user], item_code_id[item], 
                           self.estimations[user,item] if not np.isnan(self.estimations[user,item]) else 1])
        return result

# toy example
toy_train_read=pd.read_csv('./Datasets/toy-example/train.csv', sep='\t', header=None, names=['user', 'item', 'rating', 'timestamp'])
toy_test_read=pd.read_csv('./Datasets/toy-example/test.csv', sep='\t', header=None, names=['user', 'item', 'rating', 'timestamp'])

toy_train_ui, toy_test_ui, toy_user_code_id, toy_user_id_code, \
toy_item_code_id, toy_item_id_code = helpers.data_to_csr(toy_train_read, toy_test_read)


model=IKNN()
model.fit(toy_train_ui)

print('toy train ui:')
display(toy_train_ui.A)

print('similarity matrix:')
display(model.similarity_matrix_ii.A)

print('estimations matrix:')
display(model.estimations)

model.recommend(toy_user_code_id, toy_item_code_id)

toy train ui:

array([[3, 4, 0, 0, 5, 0, 0, 4],
       [0, 1, 2, 3, 0, 0, 0, 0],
       [0, 0, 0, 5, 0, 3, 4, 0]], dtype=int64)

similarity matrix:

array([[1.        , 0.9701425 , 0.        , 0.        , 1.        ,
        0.        , 0.        , 1.        ],
       [0.9701425 , 1.        , 0.24253563, 0.12478355, 0.9701425 ,
        0.        , 0.        , 0.9701425 ],
       [0.        , 0.24253563, 1.        , 0.51449576, 0.        ,
        0.        , 0.        , 0.        ],
       [0.        , 0.12478355, 0.51449576, 1.        , 0.        ,
        0.85749293, 0.85749293, 0.        ],
       [1.        , 0.9701425 , 0.        , 0.        , 1.        ,
        0.        , 0.        , 1.        ],
       [0.        , 0.        , 0.        , 0.85749293, 0.        ,
        1.        , 1.        , 0.        ],
       [0.        , 0.        , 0.        , 0.85749293, 0.        ,
        1.        , 1.        , 0.        ],
       [1.        , 0.9701425 , 0.        , 0.        , 1.        ,
        0.        , 0.        , 1.        ]])

estimations matrix:

array([[4.        , 4.        , 4.        , 4.        , 4.        ,
               nan,        nan, 4.        ],
       [1.        , 1.35990333, 2.15478388, 2.53390319, 1.        ,
        3.        , 3.        , 1.        ],
       [       nan, 5.        , 5.        , 4.05248907,        nan,
        3.95012863, 3.95012863,        nan]])

[[0, 20, 4.0, 30, 4.0],
 [10, 50, 3.0, 60, 3.0, 0, 1.0, 40, 1.0, 70, 1.0],
 [20, 10, 5.0, 20, 5.0]]

model=IKNN()
model.fit(train_ui)

top_n=pd.DataFrame(model.recommend(user_code_id, item_code_id, topK=10))

top_n.to_csv('Recommendations generated/ml-100k/Self_IKNN_reco.csv', index=False, header=False)

estimations=pd.DataFrame(model.estimate(user_code_id, item_code_id, test_ui))
estimations.to_csv('Recommendations generated/ml-100k/Self_IKNN_estimations.csv', index=False, header=False)

import evaluation_measures as ev
estimations_df=pd.read_csv('Recommendations generated/ml-100k/Self_IKNN_estimations.csv', header=None)
reco=np.loadtxt('Recommendations generated/ml-100k/Self_IKNN_reco.csv', delimiter=',')

ev.evaluate(test=pd.read_csv('./Datasets/ml-100k/test.csv', sep='\t', header=None),
            estimations_df=estimations_df, 
            reco=reco,
            super_reactions=[4,5])

943it [00:00, 12783.59it/s]

	RMSE	MAE	precision	recall	F_1	F_05	precision_super	recall_super	NDCG	mAP	MRR	LAUC	HR	Reco in test	Test coverage	Shannon	Gini
0	1.018363	0.808793	0.000318	0.000108	0.00014	0.000189	0.0	0.0	0.000214	0.000037	0.000368	0.496391	0.003181	0.392153	0.11544	4.174741	0.965327

import imp
imp.reload(ev)

import evaluation_measures as ev
dir_path="Recommendations generated/ml-100k/"
super_reactions=[4,5]
test=pd.read_csv('./Datasets/ml-100k/test.csv', sep='\t', header=None)

ev.evaluate_all(test, dir_path, super_reactions)

943it [00:00, 11968.82it/s]
943it [00:00, 12116.70it/s]
943it [00:00, 12439.74it/s]
943it [00:00, 11821.76it/s]
943it [00:00, 12771.20it/s]
943it [00:00, 12128.51it/s]
943it [00:00, 10501.52it/s]
943it [00:00, 11386.54it/s]

	Model	RMSE	MAE	precision	recall	F_1	F_05	precision_super	recall_super	NDCG	mAP	MRR	LAUC	HR	Reco in test	Test coverage	Shannon	Gini
0	Self_TopPop	2.508258	2.217909	0.188865	0.116919	0.118732	0.141584	0.130472	0.137473	0.214651	0.111707	0.400939	0.555546	0.765642	1.000000	0.038961	3.159079	0.987317
0	Ready_Baseline	0.949459	0.752487	0.091410	0.037652	0.046030	0.061286	0.079614	0.056463	0.095957	0.043178	0.198193	0.515501	0.437964	1.000000	0.033911	2.836513	0.991139
0	Self_GlobalAvg	1.125760	0.943534	0.061188	0.025968	0.031383	0.041343	0.040558	0.032107	0.067695	0.027470	0.171187	0.509546	0.384942	1.000000	0.025974	2.711772	0.992003
0	Self_TopRated	1.244026	1.000400	0.061188	0.025968	0.031383	0.041343	0.040558	0.032107	0.067695	0.027470	0.171187	0.509546	0.384942	1.000000	0.025974	2.711772	0.992003
0	Ready_Random	1.518962	1.216659	0.046978	0.019905	0.023281	0.030951	0.029399	0.019829	0.052066	0.021101	0.129283	0.506463	0.314952	0.987911	0.184704	5.110269	0.905724
0	Self_BaselineIU	0.958136	0.754051	0.000954	0.000188	0.000298	0.000481	0.000644	0.000223	0.001043	0.000335	0.003348	0.496433	0.009544	0.699046	0.005051	1.945910	0.995669
0	Self_BaselineUI	0.967585	0.762740	0.000954	0.000170	0.000278	0.000463	0.000644	0.000189	0.000752	0.000168	0.001677	0.496424	0.009544	0.600530	0.005051	1.803126	0.996380
0	Self_IKNN	1.018363	0.808793	0.000318	0.000108	0.000140	0.000189	0.000000	0.000000	0.000214	0.000037	0.000368	0.496391	0.003181	0.392153	0.115440	4.174741	0.965327

Ready-made KNNs - Surprise implementation

I-KNN - basic

import helpers
import surprise as sp
import imp
imp.reload(helpers)

sim_options = {'name': 'cosine',
              'user_based': False}  # compute similarities between items
algo = sp.KNNBasic(sim_options=sim_options)

helpers.ready_made(algo, reco_path='Recommendations generated/ml-100k/Ready_I-KNN_reco.csv',
          estimations_path='Recommendations generated/ml-100k/Ready_I-KNN_estimations.csv')

Computing the cosine similarity matrix...
Done computing similarity matrix.
Generating predictions...
Generating top N recommendations...
Generating predictions...

U-KNN - basic

import helpers
import surprise as sp
import imp
imp.reload(helpers)

sim_options = {'name': 'cosine',
              'user_based': True}  # compute similarities between users
algo = sp.KNNBasic(sim_options=sim_options)

helpers.ready_made(algo, reco_path='Recommendations generated/ml-100k/Ready_U-KNN_reco.csv',
          estimations_path='Recommendations generated/ml-100k/Ready_U-KNN_estimations.csv')

Computing the cosine similarity matrix...
Done computing similarity matrix.
Generating predictions...
Generating top N recommendations...
Generating predictions...

I-KNN - on top baseline

import helpers
import surprise as sp
import imp
imp.reload(helpers)

sim_options = {'name': 'cosine',
              'user_based': False}  # compute similarities between items
algo = sp.KNNBaseline()

helpers.ready_made(algo, reco_path='Recommendations generated/ml-100k/Ready_I-KNNBaseline_reco.csv',
          estimations_path='Recommendations generated/ml-100k/Ready_I-KNNBaseline_estimations.csv')

Estimating biases using als...
Computing the msd similarity matrix...
Done computing similarity matrix.
Generating predictions...
Generating top N recommendations...
Generating predictions...

project task 4: use a version of your choice of Surprise KNNalgorithm

# read the docs and try to find best parameter configuration (let say in terms of RMSE)
# https://surprise.readthedocs.io/en/stable/knn_inspired.html##surprise.prediction_algorithms.knns.KNNBaseline
# the solution here can be similar to examples above
# please save the output in 'Recommendations generated/ml-100k/Self_KNNSurprisetask_reco.csv' and
# 'Recommendations generated/ml-100k/Self_KNNSurprisetask_estimations.csv'

import helpers
import surprise as sp
import imp
imp.reload(helpers)

sim_options = {'name': 'cosine',
              'user_based': False}  # compute similarities between items
#i=40
#while i<=60:
#    i += 2
#    algo = sp.KNNWithMeans(k=i)
#
#    helpers.ready_made(algo, reco_path='Recommendations generated/ml-100k/Self_KNNSurprisetaskMaxk' + str(i) + '_reco.csv',
#              estimations_path='Recommendations generated/ml-100k/Self_KNNSurprisetaskMaxk' + str(i) + '_estimations.csv')

#    #best max k = 48 with RMSE: 0.955432
    
#i=1
#while i<=20:
#    i += 1
#    algo = sp.KNNWithMeans(k=48, min_k=i)
#
#    helpers.ready_made(algo, reco_path='Recommendations generated/ml-100k/Self_KNNSurprisetaskMaxk48Mink' + str(i) + '_reco.csv',
#              estimations_path='Recommendations generated/ml-100k/Self_KNNSurprisetaskMaxk48Mink' + str(i) + '_estimations.csv')

#    #best max k = 48 and min_k = 3 with RMSE: 0.954453

algo = sp.KNNWithMeans(k=48, min_k=3)

helpers.ready_made(algo, reco_path='Recommendations generated/ml-100k/Self_KNNSurprisetask_reco.csv',
        estimations_path='Recommendations generated/ml-100k/Self_KNNSurprisetask_estimations.csv')

Computing the msd similarity matrix...
Done computing similarity matrix.
Generating predictions...
Generating top N recommendations...
Generating predictions...

import imp
imp.reload(ev)

import evaluation_measures as ev
dir_path="Recommendations generated/ml-100k/"
super_reactions=[4,5]
test=pd.read_csv('./Datasets/ml-100k/test.csv', sep='\t', header=None)

ev.evaluate_all(test, dir_path, super_reactions)

943it [00:00, 10869.42it/s]
943it [00:00, 11819.04it/s]
943it [00:00, 12273.94it/s]
943it [00:00, 12777.35it/s]
943it [00:00, 12124.72it/s]
943it [00:00, 12600.44it/s]
943it [00:00, 11819.00it/s]
943it [00:00, 10867.15it/s]
943it [00:00, 12602.53it/s]
943it [00:00, 12945.89it/s]
943it [00:00, 11386.97it/s]
943it [00:00, 12128.36it/s]

	Model	RMSE	MAE	precision	recall	F_1	F_05	precision_super	recall_super	NDCG	mAP	MRR	LAUC	HR	Reco in test	Test coverage	Shannon	Gini
0	Self_TopPop	2.508258	2.217909	0.188865	0.116919	0.118732	0.141584	0.130472	0.137473	0.214651	0.111707	0.400939	0.555546	0.765642	1.000000	0.038961	3.159079	0.987317
0	Ready_Baseline	0.949459	0.752487	0.091410	0.037652	0.046030	0.061286	0.079614	0.056463	0.095957	0.043178	0.198193	0.515501	0.437964	1.000000	0.033911	2.836513	0.991139
0	Self_GlobalAvg	1.125760	0.943534	0.061188	0.025968	0.031383	0.041343	0.040558	0.032107	0.067695	0.027470	0.171187	0.509546	0.384942	1.000000	0.025974	2.711772	0.992003
0	Self_TopRated	1.244026	1.000400	0.061188	0.025968	0.031383	0.041343	0.040558	0.032107	0.067695	0.027470	0.171187	0.509546	0.384942	1.000000	0.025974	2.711772	0.992003
0	Ready_Random	1.518962	1.216659	0.046978	0.019905	0.023281	0.030951	0.029399	0.019829	0.052066	0.021101	0.129283	0.506463	0.314952	0.987911	0.184704	5.110269	0.905724
0	Self_KNNSurprisetask	0.954453	0.753598	0.039449	0.017706	0.021027	0.027213	0.035730	0.029334	0.034946	0.010791	0.069045	0.505326	0.288441	0.653977	0.084416	3.351793	0.984540
0	Ready_I-KNN	1.030386	0.813067	0.026087	0.006908	0.010593	0.016046	0.021137	0.009522	0.024214	0.008958	0.048068	0.499885	0.154825	0.402333	0.434343	5.133650	0.877999
0	Ready_I-KNNBaseline	0.935327	0.737424	0.002545	0.000755	0.001105	0.001602	0.002253	0.000930	0.003444	0.001362	0.011760	0.496724	0.021209	0.482821	0.059885	2.232578	0.994487
0	Ready_U-KNN	1.023495	0.807913	0.000742	0.000205	0.000305	0.000449	0.000536	0.000198	0.000845	0.000274	0.002744	0.496441	0.007423	0.602121	0.010823	2.089186	0.995706
0	Self_BaselineIU	0.958136	0.754051	0.000954	0.000188	0.000298	0.000481	0.000644	0.000223	0.001043	0.000335	0.003348	0.496433	0.009544	0.699046	0.005051	1.945910	0.995669
0	Self_BaselineUI	0.967585	0.762740	0.000954	0.000170	0.000278	0.000463	0.000644	0.000189	0.000752	0.000168	0.001677	0.496424	0.009544	0.600530	0.005051	1.803126	0.996380
0	Self_IKNN	1.018363	0.808793	0.000318	0.000108	0.000140	0.000189	0.000000	0.000000	0.000214	0.000037	0.000368	0.496391	0.003181	0.392153	0.115440	4.174741	0.965327