Adaptive neuro fuzzy network for recommendation

y o c u -tr a c k c ADAPTIVE NEURO-FUZZY NETWORK FOR RECOMMENDATION In Partial Fulfillment of the Requirements of the Degree of MASTER OF INFORMATION TECHNOLOGY MANAGEMENT In Computer Science and Engineer By Mr Nguyen Duc Anh ID: MITM05001 International University - Vietnam National University HCMC March 2015 d o m o w w w d o C lic k to bu y bu to k lic C w w w N O W ! h a n g e Vi e N PD ! XC er O W F- w m h a n g e Vi e w PD XC er F- c u -tr a c k c h a n g e Vi e w N y bu to k lic c u -tr a c k ADAPTIVE NEURO-FUZZY NETWORK FOR RECOMMENDATION d o m w o c C m o d o w w w w w C lic k to bu y N O W ! XC er O W F- w PD h a n g e Vi e ! XC er PD F- c u -tr a c k In Partial Fulfillment of the Requirements of the Degree of MASTER OF INFORMATION TECHNOLOGY MANAGEMENT In Computer Science and Engineer By Mr Nguyen Duc Anh ID: MITM05001 International University - Vietnam National University HCMC March 2015 Under the guidance and approval of the committee, and approved by all its members, this thesis has been accepted in partial fulfillment of the requirements for the degree Approved: Chairperson Committee member Committee member Committee member Committee member Committee member ii c y o c u -tr a c k c Acknowledgments Throughout my thesis and development process, it is impossible for me to well complete all my tasks and missions without the support and encouragement from the other ones At first, I would like to thank Dr Duong Trong Hai He is always by my side to support me identify the main ideas of this research this is the most important support for me He instructs me to be familiar with data-mining, machine-learning, etc Moreover, He is willing to give me helpful advices whenever I have difficulties or troubles with my thesis I am grateful to my family, who encourages and motivates me to keep moving forward There are also my colleagues, schoolmates who also support and help me directly and indirectly; I want to say thank all of them iii d o m o w w w d o C lic k to bu y bu to k lic C w w w N O W ! h a n g e Vi e N PD ! XC er O W F- w m h a n g e Vi e w PD XC er F- c u -tr a c k c h a n g e Vi e w N y bu to k lic c u -tr a c k Plagiarism Statements I would like to declare that, apart from the acknowledged references, this thesis either does not use language, ideas, or other original material from anyone; or has not been previously submitted to any other educational and research programs or institutions I fully understand that any writings in this thesis contradicted to the above statement will automatically lead to the rejection from the MITM program at the International University – Vietnam National University Hochiminh City iv d o m w o c C m o d o w w w w w C lic k to bu y N O W ! XC er O W F- w PD h a n g e Vi e ! XC er PD F- c u -tr a c k c h a n g e Vi e w N y bu to k lic c u -tr a c k Copyright Statement This copy of the thesis has been supplied on condition that anyone who consults it is understood to recognize that its copyright rests with its author and that no quotation from the thesis and no information derived from it may be published without the author’s prior consent © Nguyen Duc Anh/MITM05001/2012-2015 v w d o o c m C m o d o w w w w w C lic k to bu y N O W ! XC er O W F- w PD h a n g e Vi e ! XC er PD F- c u -tr a c k c h a n g e Vi e w N y bu to k lic c u -tr a c k Plagiarism Statements iv Copyright Statement v This Thesis based on Publications x Abstract xi Chapter 1: Introduction 1.1 Motivation 1.2 Goals of the Dissertation 1.3 Overal Approach 1.4 Related Work 1.5 Thesis Outline Chapter 2: User Behaviors-based CF Using Neuro-Fuzzy Network 2.1 Profile Modeling 2.2 Content-based Filtering Using Neuro-Fuzzy Network Chapter Experiments 12 3.1 Dataset Introduction 12 3.1.1 Overview 12 3.1.2 Dataset analysis 13 3.2 Applied ANFIS to netflix dataset 14 3.2.1 ANFIS Model 14 3.2.2 Run the testing dataset 25 3.3 Evaluation Methods 27 3.4 Practice and Result 27 3.4.1 Movie 329 28 3.4.2 Movie 30 30 3.4.3Movie 2464 31 vi w d o m Table of Contents o c C m o d o w w w w w C lic k to bu y N O W ! XC er O W F- w PD h a n g e Vi e ! XC er PD F- c u -tr a c k c h a n g e Vi e w N y bu to k lic c u -tr a c k 3.4.4 Movie 2848 33 3.4.5Movie 2548 34 3.5 Evaluation results 36 Chapter 4: Conclusion 38 References 39 vii d o m w o c C m o d o w w w w w C lic k to bu y N O W ! XC er O W F- w PD h a n g e Vi e ! XC er PD F- c u -tr a c k c h a n g e Vi e w N y bu to k lic c u -tr a c k Fg 2.1.1 Profile generation process Fg3.1.1 Netflix Dataset structure 12 Fg3.1.2 Rating-scores statistic 13 Fg3.1.2 The rating-scores comparison for top 10 movies have highest number of rating14 Fg3.2.1 The ANFIS’s structure 14 Fg3.2.1 The main workflow of ANFIS 15 Fg3.2.1 3Sample of user profile Level 16 Fg3.2.1 Sample of user profile Level 18 Fg3.2.1 5HyperBox dataset and PureBox Dataset where before and after clustered by NCP 18 Fg3.2.1 Samples of purebox clusters 20 Fg3.2.1 The Max-Min PureBox 21 Fg3.2.1 The final result of the user profile building steps 22 Fg3.2.1 Samples of data use to training by Perceptron 23 Fg3.3 Distribution of Tranning set and Testing set in dataset 28 Fg3.4.2 Comparison between training data set and real dataset of movie 30 30 Fg3.4.2 Result of 100 samples used to test for movie 30 31 Fg3.4.3 Comparison between training data set and real dataset of movie 2464 32 Fg3.4.3 Result of 100 samples used to test for movie 2464 32 Fg3.4.4 Comparison between training data set and real dataset of movie 2848 33 Fg3.4.4 Result of 100 samples used to test for movie 2848 34 Fg3.4.5 Comparison between training data set and real dataset of movie 2548 35 Fg3.4.5 Result of 100 samples used to test for movie 2548 35 Fg3.5 MAE and RMSE of movies 2464,2548,30,2848,329 36 viii w d o m List of Figures o c C m o d o w w w w w C lic k to bu y N O W ! XC er O W F- w PD h a n g e Vi e ! XC er PD F- c u -tr a c k c h a n g e Vi e w N y bu to k lic c u -tr a c k Table 3.2.1 Samples of W had computed by Perception for Movie 329 23 Table 3.2.2.1Predict Rating-scores for userssamples, movie 329 26 Table 3.4.1.1 Comparison between training data set and real dataset of movie 329 28 Table3.4.2 Comparison between training data set and real dataset of movie 30 30 Table3.4.3 Comparison between training data set and real dataset of movie 2464 31 Table3.4.4 Comparison between training data set and real dataset of movie 2848 33 Table3.4.5 Comparison between training data set and real dataset of movie 2548 34 ix w d o m List of Table o c C m o d o w w w w w C lic k to bu y N O W ! XC er O W F- w PD h a n g e Vi e ! XC er PD F- c u -tr a c k c h a n g e Vi e w N y bu to k lic c u -tr a c k This Thesis based on Publications International Conference Publications (Accepted) Duc Anh Nguyen and Trong Hai Duong, “Video Recommendation Using NeuroFuzzy on Social TV Environment”, International conference on Computer Science, Applied Mathematics and Applications (ICCSAMA 2015) published in a volume of series Advances in Intelligent Systems and Computing of Springer Verlag, indexed by ISI Proceedings, DBLP, Ulrich's, EI-Compendex, SCOPUS, Zentralblatt Math, MetaPress, Springerlink Issues in ISI-SCI journals International Journal Publications (Submitted) Trong Hai Duong and Duc Anh Nguyen, “User Behaviors-based Collaborative Filtering for Video Recommendation Using Ontology-based Neuro-Fuzzy on Social TV”, ELSEVIER, 03-2015 x w d o o c m C m o d o w w w w w C lic k to bu y N O W ! XC er O W F- w PD h a n g e Vi e ! XC er PD F- c u -tr a c k c y o c u -tr a c k c Fg3.4.1 Comparison between training data set and real dataset of movie 329 In this case, I set the maximum loops for the perception is 1000, it means the W which is shown in Table3.2.1.1 is the best W in 1000 loops of the perception Fg3.4.1 Result of 100 samples used to test for movie 329 With movie 329, the data of rating labels used for training dataset are equality and the result is quite good, at some points, the chart in Fig 3.4.1.2 still has some big distances between Prediction Value and the Real Value 29 d o m o w w w d o C lic k to bu y bu to k lic C w w w N O W ! h a n g e Vi e N PD ! XC er O W F- w m h a n g e Vi e w PD XC er F- c u -tr a c k c y o c u -tr a c k c d o m o w w w d o C lic k to bu y bu to k lic C w w w N O W ! h a n g e Vi e N PD ! XC er O W F- w m h a n g e Vi e w PD XC er F- c u -tr a c k 3.4.2 Movie 30 The training dataset of movie 30 is collected randomly; just ensure 1000 users are collected Movie Description Rating Rating Rating Rating Rating Training Set 16 69 254 434 227 Real Dataset 2,555 7,594 31,521 50,570 26,173 30 Table3.4.2 Comparison between training data set and real dataset of movie 30 Fg3.4.2 Comparison between training data set and real dataset of movie 30 30 c y o c u -tr a c k c d o m o w w w d o C lic k to bu y bu to k lic C w w w N O W ! h a n g e Vi e N PD ! XC er O W F- w m h a n g e Vi e w PD XC er F- c u -tr a c k Fg3.4.2 Result of 100 samples used to test for movie 30 With movie’ id 30, the samples with rating-scores and are very small in comparison with samples with rating-scores 3, and 5, thus the model have wrong prediction for tests with rating-scores and 2, shown in Fig: 3.4.2.1 3.4.3Movie 2464 For the case study of movie 2464, we randomly collect 200 samples for each rating label (1-5) As shown in Table 3.4.3.1 and Fig 3.4.3.1 Movie Description Rating Rating Rating Rating Rating Training Set 200 200 200 200 200 Real Dataset 221 498 2073 2374 983 2464 Table3.4.3 Comparison between training data set and real dataset of movie 2464 31 c y o c u -tr a c k c Fg3.4.3 Comparison between training data set and real dataset of movie 2464 Fg3.4.3 Result of 100 samples used to test for movie 2464 With movie 2464, the predicted value is more closed with the real value in comparison with movie 30 The reason here is that rating-scores 1,2,3,4 and of 32 d o m o w w w d o C lic k to bu y bu to k lic C w w w N O W ! h a n g e Vi e N PD ! XC er O W F- w m h a n g e Vi e w PD XC er F- c u -tr a c k c y o c u -tr a c k c d o m o w w w d o C lic k to bu y bu to k lic C w w w N O W ! h a n g e Vi e N PD ! XC er O W F- w m h a n g e Vi e w PD XC er F- c u -tr a c k samples in the training set used for the movie 2464 is significant balance to be train for the ANFIS Therefore, the test result of these movies is better 3.4.4 Movie 2848 For the case study of movie 2848, we randomly collect 1000 samples in total As shown in Table 3.4.4.1 and Fig 3.4.4.1 Movie Description Rating Rating Rating Rating Rating Training Set 96 303 132 274 193 Real Dataset 166 551 3435 7825 5424 2848 Table3.4.4 Comparison between training data set and real dataset of movie 2848 Fg3.4.4 Comparison between training data set and real dataset of movie 2848 33 c y o c u -tr a c k c d o m o w w w d o C lic k to bu y bu to k lic C w w w N O W ! h a n g e Vi e N PD ! XC er O W F- w m h a n g e Vi e w PD XC er F- c u -tr a c k Fg3.4.4 Result of 100 samples used to test for movie 2848 The samples in the training set used for the movie 2848 are significant balance and much enough to be trained for the ANFIS Therefore, the test result of this movie is better than movie 30 3.4.5Movie 2548 For the case study of movie 2548, we randomly collect 1000 samples in total As shown in Table 3.4.5.1 and Fig 3.4.5.1 Movie Description Rating Rating Rating Rating Rating Training Set 255 145 66 153 381 Real Dataset 470 281 719 1998 5321 2548 Table3.4.5 Comparison between training data set and real dataset of movie 2548 34 c y o c u -tr a c k c Fg3.4.5 Comparison between training data set and real dataset of movie 2548 Fg3.4.5 Result of 100 samples used to test for movie 2548 The best result is the case study of the movie 2548, where the rating-scores of these samples are more balance than afore mentioned samples for movies 30, 2464, 329 and 2848 35 d o m o w w w d o C lic k to bu y bu to k lic C w w w N O W ! h a n g e Vi e N PD ! XC er O W F- w m h a n g e Vi e w PD XC er F- c u -tr a c k c y o c u -tr a c k c 3.5 Evaluation results We used 100 testing samples corresponding to movies 30, 329, 2464, 2548, 2848 The comparisons of MAE and RMSE between the movies are shown in Fig: 3.5.1 With movie 30, the samples with rating-scores and are very small in comparison with samples with rating-scores 3, and 5, thus the model have wrong prediction for tests with rating-scores and Furthermore, for both studying of movie id 30 and 329, the samples are very small in comparison with the real dataset (only 0,8% and 1,5% of the real dataset) thus the model is wrong prediction for tests Therefore, the difference between MAE and RMSE of movie 30 and 329 are significant since distance between real rating and predicted rating of testing samples Fg3.5 MAE and RMSE of movies 2464,2548,30,2848,329 With movies 2464 and 2848, the predicted value is closer with the real value in comparison with movie 30 The reason here is that rating-scores 1,2,3,4 and 36 d o m o w w w d o C lic k to bu y bu to k lic C w w w N O W ! h a n g e Vi e N PD ! XC er O W F- w m h a n g e Vi e w PD XC er F- c u -tr a c k c y o c u -tr a c k c of samples in the training set used for movies 2464, 2848 are significant balance to train for the ANFIS Therefore, the test result of these movies is better The best result is the case study of the movie 2548, where the ratingscores of these samples are more balance than aforementioned samples’ for movies 30, 329, 2464 and 2848.From these experimental results, we can conclude that the sample’s rating-scores are more balance and significant to compare with the real data, the results are more accurate 37 d o m o w w w d o C lic k to bu y bu to k lic C w w w N O W ! h a n g e Vi e N PD ! XC er O W F- w m h a n g e Vi e w PD XC er F- c u -tr a c k c y o c u -tr a c k c Chapter 4: Conclusion The idea of this proposed method is to adjust users’ social web behavior to their own ratings dual with a target video In particular, a user profile is learned by the user’s social web behavior This user profile is presented by a vector For each target video, we collect all users’ profiles who rated on the target video Each user’s profile are considered as an input vector and his/her corresponding rating-score is a output value of the fuzzy neural network The trained neural network is used to predict the rating of a user to the target video We use Netflix data-set to evaluate the proposed method The result shown that the proposed method is a significant effective approach However, the result will go downhill if the rating-scores of training set are not balance and significant 38 d o m o w w w d o C lic k to bu y bu to k lic C w w w N O W ! h a n g e Vi e N PD ! XC er O W F- w m h a n g e Vi e w PD XC er F- c u -tr a c k c y o c u -tr a c k c References Aroyo, L N., Lyndon and Dietze, S.: 2009, Television and the future internet: the No-Tube project In: Future Internet Symposium (FIS) 2009, 1-3 September 2009, Berlin,Germany Nguyen,S D., Ngo, K N.: 2008, An Adaptive Input Data Space Parting Solution to theSynthesis of Neuro-Fuzzy Models, International Journal of Control, Automation, andSystems, IJCAS, Vol 6, No 6, 928-938 Nguyen, S D., Choi, S B.: 2012,Anew Neuro-Fuzzy Training Algorithm for IdentifyingDynamic Characteristics of Smart Dampers, Smart Materials and Structures, Vol 21 Pazzani, M.J., Billsus, D.: 2007, Content-based recommendation systems, in: P Brusilovsky, A Kobsa,W Nejdl (Eds.), The AdaptiveWeb, Lecture Notes in ComputerScience, vol 4321, Springer-Verlag, 2007, pp 325?341 Resnick, P., Iacovou, N., Suchack, M., Bergstrom, P., Riedl, J.T.: 1994, GroupLens: anopen architecture for collaborative filtering of netnews, in: Proceedings of the ACMConference on Computer Supported CooperativeWork, 1994, pp 175-186 Breese, J.S., Heckerman, D., Kadie, C.: 1999, Empirical analysis of predictive algorithmsfor collaborative filtering, in: Proceedings of the 14th Conference on Uncertainty inArtificial Intelligence, 1999, pp 43-52 Goldberg, K., Roeder, T., Gupta, D., Perkins, C.: 2001, Eigentaste: a constant time collaborative filtering algorithm, Information Retrieval, (2), 2001, pp 133-151 39 d o m o w w w d o C lic k to bu y bu to k lic C w w w N O W ! h a n g e Vi e N PD ! XC er O W F- w m h a n g e Vi e w PD XC er F- c u -tr a c k c y o c u -tr a c k c Smeaton, A F., Over, P., and Kraaij, W.: 2006, Evaluation campaigns and TRECVid,? InProc of the ACM MIR, 2006, pp 321-330 Barragns-Martnez, A.B., Costa-Montenegro, E., Burguillo-Rial, J.C., ReyLpez, M.,Mikic-Fonte, F.A and Peleteiro-Ramallo, A.: A hybrid content-based and item-basedcollaborative filtering approach to recommend TV programs enhanced with singularvalue decomposition, ;presented at Inf Sci., 2010, pp.4290-4311 10 Li, Q.,Wang, J., Chen, Y.P., and Lin, Z.: 2010, User comments for newsrecommendationin forum-based social media, ;presented at Inf Sci., 2010, pp.4929-4939 11 Baeza-Yates, R., Ribeiro-Neto, B.: 1999, Modern information retrieval, Addison WesleyLongman Publisher, 1999 12 Brin, S., Motwani, R., Page, L., Winograd, T.: 1998, What can you with a Web inyour pocket, in: Bulletin of the IEEE Computer Society TechnicalCommittee on DataEngineering, 1998, pp 37-47 13 Del Corso, G.M., Gull, A., Romani, F.: 2005, Ranking a stream of news, in: Proceedingsof the 14th International Conference onWorldWideWeb (WWW), pp 97-106 14 Lavrenko, V., Schmill, M., Lawrie, D., Ogilvie, P., Jensen, D., Allan, J.: 2000, Languagemodels for financial news recommendation, in: Proceedings of the Ninth InternationalConference on Information and Knowledge Management (CIKM), pp 389-396 40 d o m o w w w d o C lic k to bu y bu to k lic C w w w N O W ! h a n g e Vi e N PD ! XC er O W F- w m h a n g e Vi e w PD XC er F- c u -tr a c k c y o c u -tr a c k c 15 Ardissono, L., Gena, C., Torasso, P., Bellifemine, F., Difino, A., Negro, B.: 2004, Usermodeling and recommendation techniques for personalized electronic program guides,in: Personalized Digital Television-Targeting Programs to Individual Viewers, Human-Computer Interaction Series, vol 6, Kluwer Academic Publishers, pp 3-26 (Chapter1) 16 Baudisch, P., Brueckner, L.: 2002, TV scout: lowering the entry barrier to personalizedTVprogram recommendation, in: Bra P.D , Brusilovsky P., Conejo R (Eds.), Proceedingsof Adaptive Hypermedia and AdaptiveWeb-Based Systems (AH?02), Lecture Notes inComputer Science, vol 2347, Springer, pp 58-68 17 Golub, G., Loan, C.V.: 1996, Matrix Computations, third ed., Johns Hopkins Studies inMathematical Sciences, Baltimore 18 Duong, T.H., Uddin, M N., Li, D., Jo, G.S.: 2009, A Collaborative Ontology-Based UserProfiles System, ICCCI 2009, Lecture Notes in Computer Science, 540-552 19 Netflix Prize: Forum/ Dataset README flie, http://www.netflixprize.com/, Editor 2006 20 Netflix, netflix dataset N Prize, Editor 2011, lifecrunch.biz 21 KANOKWAN.K.,CHUN.C.F.: 2012, Neural Network Modeling for an Intelligent Recommendation System Supporting SRM for Universities in Thailand, Vol.11, E-ISSN: 2224-2872 22 Michael.R.S, Tony.M, Michael.G.:2014, A Hybrid Latent Variable Neural Network Model for Item Recommendation 41 d o m o w w w d o C lic k to bu y bu to k lic C w w w N O W ! h a n g e Vi e N PD ! XC er O W F- w m h a n g e Vi e w PD XC er F- c u -tr a c k c y o c u -tr a c k c 23 Jianfeng G, Patrick P, Michael G, Xiaodong H, Li D: 2014, Modeling Interestingness with Deep Neural Networks, EMNLP 2014,Microsoft Research, pp 2-13 24 Christina.C, Andreas.S.: 2005, A Hybrid Movie Recommender System Based on Neural Networks, ISDA’05, IEEE 42 d o m o w w w d o C lic k to bu y bu to k lic C w w w N O W ! h a n g e Vi e N PD ! XC er O W F- w m h a n g e Vi e w PD XC er F- c u -tr a c k c y c d o m w o o c u -tr a c k w w d o C lic k to bu y bu to k lic C w w w N O W ! h a n g e Vi e N PD ! XC er O W F- w m h a n g e Vi e w PD XC er F- c u -tr a c k c [...]... c In fact, there are many researches on recommendation systems One of them is the research named: “Neural Network Modeling for an Intelligent Recommendation System Supporting SRM for Universities in Thailand”, [21] proposed by Kanokwan Kongsakun and Chun Che Fung This is a recommendation system proposal, used to predict and recommend the appropriate courses for students thereby increase their chance... f: (1) | =f( ) In this paper, the above mathematical model is expressed by a fuzzy- neuron structure (FNS), which is a combination of a fuzzy inference system (FIS) and a neural network structure (NNS) Relating to the FIS, it can be summarized as follows The FIS is built based on the algorithm establishing an adaptive neuro- fuzzy system, ENFS [3] By using data driven method, the same features or characteristics... authors used Neural Network techniques to find the structures and relationships within data and final GPA of freshmen in subjects of interest The authors [21] had come to the conclusion that recommendation system is a useful service According to another research named: “A Hybrid Latent Variable Neural Network Model for Item Recommendation [22] The authors [22] proposed neural network model with latent...h a n g e Vi e w N y bu to k lic c u -tr a c k Abstract Recommendation systems are systems that seek for prediction and give users recommendation about products or items that they might be interested in There are two common approaches, which have been proposed to perform recommendation system; they are content-based filtering (CBF) and collaborative filtering (CF)... this work, we consider recommendation on the context of Social TV (STV) The watchers/users may either share, comment, rate, or tag videos in which they are interested in Each video must be watched and rated by many users For these assumptions, we proposed a novel model-based collaborative filtering using a fuzzy neural network to learn user’s social web behaviors to make video recommendation on STV... part is chapter 2 named “User Behaviors-based CF Using NeuroFuzzy Network The main purpose of this part is to analyze in detail the relevant theories such as User modeling, ANFIS, TF/IDF, Perceptron, etc which will apply in my thesis research paper Chapter 3 is Experiment This chapter introduces about applying the proposed novel ANFIS for Video recommendation system and introduces the 5 d o m o w w... n g e Vi e N PD ! XC er O W F- w m h a n g e Vi e w PD XC er F- c u -tr a c k c y o c u -tr a c k c Chapter 2: User Behaviors-based CF Using Neuro- Fuzzy Network 2.1 Profile Modeling User profile can be static and dynamic information In static, user permanent information such as name, age, sex, educational background etc is included, whereas the dynamic user profile, the less permanent characteristics... k c y o c u -tr a c k c trained neural network is used to predict the rating of a user to the target video We use netflix data set to evaluate the proposed method 1.4 Related Work The trend for using online social networks to talk about TV programs and to share their opinions with others, is increasing This reflected with the dissemination of platforms designed for Social TV [1] The NoTube [1] brings... significant effective method Keywords: ANFIS, Ontology, Smart TV, Video, Recommendation system, and Neural network xi w d o o c m C m o d o w w w w w C lic k to bu y N O W ! XC er O W F- w PD h a n g e Vi e ! XC er PD F- c u -tr a c k c y o c u -tr a c k c Chapter 1: Introduction 1.1 Motivation Recommendation is a subclass of information filtering, which uses data on past user preferences to predict... are no any neighbors, the traditional CF’s result is gone downhill To solve the aforementioned problem, we proposed a novel model-based collaborative filtering using a fuzzy neural network to learn user’s social web behaviors for video recommendation on STV 1.2 Goals of the Dissertation Our goals in this thesis focused on solve the problem of lack of neighbors in the traditional CF In that, we predict