Kaggle kenneth

WSDM Challenge
Recommender Systems
Kenneth Emeka Odoh
25 Jan, 2018 ( Kaggle Meetup | SFU Ventures Labs )
Vancouver, BC

Talk Plan
● Bio
● Competition
● Winning Solutions
● My solution
● Conclusions
● References
2
Kenneth Emeka Odoh

Bio
Education
● Masters in Computer Science (University of Regina) 2013 - 2016
Work
● Research assistant in the field of Visual Analytics 2013 - 2016
○ ( https://scholar.google.com/citations?user=3G2ncgwAAAAJ&hl=en )
● Software engineer working in an IOT startup (Current position) 2017 -
3Kenneth Emeka Odoh

Competition
● Competition rule
○ Only 5 submissions per day
● Data Exploration
○ Data visualization
● Evaluation
● Primer on recommender systems
4
Kenneth Emeka Odoh

Nature of Data
● Data of Listening habits.
● Data are chronologically ordered.
○ Implicit time series data without
timestamps.
○ Timestamps are excluded for sake of
avoiding data leakages.
● Balanced data sets.
What to predict
● Given a data set of listening habit of
users.
● If a song has been listened in recent
past, predict if it will be listening to
again in the same month.
● If the song is listened to again by the
same user in the same month, output
1 or 0 otherwise
Competition: Data Exploration
5
Kenneth Emeka Odoh

Competition: Background
● URL ( https://www.kaggle.com/c/kkbox-music-recommendation-challenge )
● Sister competition on churn prediction by same company.
● Total price $5000
Class of Problem
● Classification
● Sequence prediction ? 6Kenneth Emeka Odoh

Competition: Data files
● train.csv
○ msno, song_id, source_system_tab, source_screen_name, source_type,
target
● songs.csv
○ song_id, song_length, genre_ids, artist_name, composer, lyricist,
language
● members.csv
○ msno, city, bd, gender, registered_via, registration_init_time,
expiration_date
● song_extra_info.csv
○ song_id, song name, isrc
7
Kenneth Emeka Odoh

Competition: Visualization of Data
8
Figure 1: Count vs Source typeKenneth Emeka Odoh
[jeru666]

9Figure 2: Count vs Source type grouped by labels
Kenneth Emeka Odoh
[jeru666]

10Figure 3: Source screen names vs Count
Kenneth Emeka Odoh
[jeru666]

11
Figure 4: Count vs Source screen names grouped by labelKenneth Emeka Odoh
[jeru666]

12
Figure 5: Count vs Source system tabKenneth Emeka Odoh
[jeru666]

13
Figure 6: Proportion by SexKenneth Emeka Odoh
[jeru666]

14
Figure 7: Proportion by Sex grouped by labelKenneth Emeka Odoh
[jeru666]

15
Figure 8: Proportion by LanguageKenneth Emeka Odoh
[jeru666]

16
Figure 9: City vs CountKenneth Emeka Odoh
[jeru666]

Competition: Evaluation (AUC)
Image credit: Walber & http://gim.unmc.edu/dxtests/roc2.htm
17
Optimizing PR space -> optimizing ROC space [Davis .et.al]
Kenneth Emeka Odoh

Evaluation Metric
● AUC per user vs AUC per group ?
● What about GAUC?
○ calculate the average of the AUC scores of each user.
■ Cold start problem
● AUC vs F-score vs RMSE ?
18Kenneth Emeka Odoh

Competition: Recommender Systems
19
Kenneth Emeka Odoh

U: set of users
I : set of items
R: set of rating of users on items.
Iu
: set of items rated by a user, u.
f : U x I→ R
This is usually a sparse matrix and as such so
form of data imputation is needed
The most common form of recommender systems
● Best-item recommendation
● Top-N recommendation
For a user, ua
, and i* has estimated the highest
value
i* = argmax f(ua
, j )
j ∊ IIu
20
Kenneth Emeka Odoh

WINNING SOLUTIONS
21
Kenneth Emeka Odoh

1st Place Solution
● Similar to Click Value Rate (CVR)
○ Click-through rate (CTR) is the ratio of users who click on a specific link to the number of total users who
accessed the resource.
● Relistening is like purchasing, and first listening is like clicking.
● Not suitable for latent-factor based or neighborhood-based collaborative filtering is not the best
way for this problem.
● Missing not at random.
● Data is sparse as a user-song pair is very informative.
● Data is time sensitive.
Kenneth Emeka Odoh
22

Ensembling
0.6 * LightGBMs and 0.4 * NNs
● Single LightGBMs is 0.74440 on public LB with a learning rate of 0.1.
● Single LightGBMs is 0.74460 on public LB with a learning rate of 0.05.
● Single LightGBMs with bagging is 0.74584 on public LB with a learning rate of 0.05.
For NNs, my best 30-ensemble on the same feature set can get 0.74215, and bagging ensemble of NNs on
different feature set can get 0.74333.
The predictions of LightGBMs and NNs are quite uncorrelated (correlation coefficient is about 0.92). The
ensemble of single LightGBM and single 30-ensemble of NN can get 0.7489+. The ensemble of ensembles
can reach 0.7498+.
Data preprocessing
https://kaggle2.blob.core.windows.net/forum-message-attachments/259372/8131/model.png
Kenneth Emeka Odoh
23

Analyzing Missing Data
Missing-data mechanisms
● Missing completely at random: This occurs when the probability of missing is same for all cases.
● Missing at random: The probability of missing depends only on available information.
● Missingness that depends on unobserved predictors
● Missingness that depends on the missing value itself.
Handling missing data
● Exclude missing data
● Mean / median imputation
● Last value carried forward
● Using information from related observation
24
Kenneth Emeka Odoh
[http://www.stat.columbia.edu/~gelman/arm/missing.pdf]

3rd Place Solution
Identified how to generate features from the future. We can listen or not listen to the same artist by
the same user in the future.
Add matrix factorization to the feature to use xgboost and catboost.
last 35% for future feature
earlier 65% for history
The size of X_train, X_val was about 20% of data.
Kenneth Emeka Odoh
25

The history can be generated using target. Other features uses all data. The generated features in the
context of categorical features.
1) mean of target from history by categorical features (pairs and triples)
2) count – the number of observations in all data by categorical features (pairs and triples)
3) regression – linear regression target by categorical features (msno, msno + genre_ids, msno+composer
and so on)
4) time_from_prev_heard – time from last heard by categorical features (msno, msno + genre_ids,
msno+composer and so on)
5) time_to_next_heard – time to next heard by categorical features (pairs and triples which include msno)
6) last_time_diff – time to the last heard by categorical features
7) part_of_unique_song – the share of unique song of artist that the user heard on all unique song of artist

8) matrix_factorization – LightFM. I use as a user msno in different context (source_type) and the msno
was the feature for user.
I use last 35% for fit. Also I drop last 5% and again use 35%. Such a procedure could be repeated many
times and this greatly improved the score (I did 6 times on validation and this gave an increase of 0.005, but
in the end I did not have time to do everything and did only 2 times on test).
As a result, I fit xgboost and catboost on two parts of the data using and without the matrix_factorization
feature. And finally I blend all 8 predictions.
The code is available here:
https://github.com/VasiliyRubtsov/wsdm_music_recommendations/blob/master/pipeline.ipynb

Factorization Machine
● By factoring into a number of latent vectors.
● lends itself to kernels.
Let us dive into factorization in a new thread.
https://www.csie.ntu.edu.tw/~r01922136/slides/ffm.pdf
Observed Characteristics of factorization machine
● Ideal for handling missing data
28
Kenneth Emeka Odoh

6th Place Solution
I use an ensemble (average) of lightgbm. The best single model public LB 0.726
Feature Engineering
1) raw features for both song, user and context
2) SVD matrix factorization latent space of user-song interaction matrix
3) SVD matrix factorization score of user-song interaction matrix
4) collaborative filtering (CF) score (top 100 neighborhoods ) of user-song interaction
5) similarity of artist, lyricist, composer, genre,language, source_system_tab, source_type,
soure_screen_name,within each user.
6) source_system_tab, source_type, soure_screen_name categorical variable rate of each user.
7) word embedding of artist, lyricist, composer, genre.
I use 10 cross validation, where train / valid set are splitted randomly.
Kenneth Emeka Odoh
29

SVD
30
[https://intoli.com/blog/pca-and-svd/]
Kenneth Emeka Odoh

Word Embeddings
31[https://deeplearning4j.org/word2vec.html]Kenneth Emeka Odoh

My Solution
● Primer on LSTM architectures.
○ Bidirectional lstm.
○ Batch normalization
● Intuition / guide for building reasonable neural architecture.
● Tricks to avoid overfitting.
● Optimizing loss function using AUC.
● Cross validation of time series data.
● Discuss failures in my traditional ML models.
○ Try same model with different custom loss functions
32
Kenneth Emeka Odoh

LSTM
● LSTM is RNN trained using BPTT ( solves
vanishing gradient problem ).
● Capture long term temporal dependency.
● A memory blocks used in place of neuron.
There are three types of gates in a memory block:
○ Forget Gate: decides information to be
removed from the unit.
○ Input Gate: decides information to update
memory state.
○ Output Gate: decides information to output
based on input and the memory state.
Goodfellow, et.al. Deep Learning, 2016
33
Kenneth Emeka Odoh

34[Fei-Fei et. al., 2017]Kenneth Emeka Odoh

35[Fei-Fei et. al., 2017]Kenneth Emeka Odoh

LSTM Architecture
iooi
[Google images]
See tutorial on subject [https://machinelearningmastery.com/timedistributed-layer-for-long-short-term-memory-networks-in-python/]
Many-to-many model (boundary detection problem)
36
Kenneth Emeka Odoh
[Fei-Fei et. al., 2017]

● Bidirectional LSTMs can be trained
using past and future data of a time
frame.
● Bidirectional LSTMs can improve
performance on a number of sequence
classification.
● Bidirectional LSTMs is two lstms
where the 1st lstm accept input as-is,
and 2nd accept the input in reverse
order.
Bidirectional LSTM
37
[Yu, Zhou et al.]
Kenneth Emeka Odoh

Batch Normalization
● Normalization in neural architecture.
● Reduce covariance shift thereby improving
training speed.
● Control gradient flow in the network
thereby minimizing saturation in non-relu
based activation.
● Fix distribution between training and
testing set thereby enhancing
generalization.
● Implicit regularization.
[Ioffe et. al 2015]
38
Kenneth Emeka Odoh

Overfitting
● Implicit Regularization
○ Batch normalization
○ Early stopping [https://en.wikipedia.org/wiki/Early_stopping]
● Dropout
39
Kenneth Emeka Odoh
[Srivastava et. al., 2014]

Cross-Validation of Time-Series
● Don’t trust the
leaderboard.
● Trust local CV for model
comparison.
40
Kenneth Emeka Odoh

My Solution: Code
def model_relu6():
model = Sequential()
model.add( Embedding(inputDim*inputDim*inputDim, inputDim, dropout=0.2) )
#input vector dimension
model.add(Convolution1D(nb_filter= nb_filter, filter_length= filter_length,
border_mode='valid', activation='relu',subsample_length=1))
model.add(MaxPooling1D(pool_length= pool_length))
model.add( Bidirectional( LSTM(1024, return_sequences=True) ))
model.add(LeakyReLU())
model.add(BatchNormalization( ))
model.add(Dropout(0.5))
model.add( Bidirectional( LSTM(2048, return_sequences=True)) )
model.add(Activation('relu'))
model.add( Bidirectional( LSTM(512, return_sequences=True)) )
model.add( Bidirectional( LSTM(256)) )
model.add(Dense(1, activation='sigmoid'))
model.compile(loss='binary_crossentropy',
optimizer='adam',metrics=[jacek_auc,discussion41015_auc])
return model
41Code: https://github.com/kenluck2001/KaggleKenneth/tree/master/wsdm_Kaggle
#full code here ( https://www.kaggle.com/rspadim/gini-keras-callback-earlystopping-validation )
# FROM https://www.kaggle.com/c/porto-seguro-safe-driver-prediction/discussion/41108
def jacek_auc(y_true, y_pred):
score, up_opt = tf.metrics.auc(y_true, y_pred)
#score, up_opt = tf.contrib.metrics.streaming_auc(y_pred, y_true)
K.get_session().run(tf.local_variables_initializer())
with tf.control_dependencies([up_opt]):
score = tf.identity(score)
return score
# FROM https://www.kaggle.com/c/porto-seguro-safe-driver-prediction/discussion/41015
# AUC for a binary classifier
def discussion41015_auc(y_true, y_pred):
ptas = tf.stack([binary_PTA(y_true,y_pred,k) for k in np.linspace(0, 1, 1000)],axis=0)
pfas = tf.stack([binary_PFA(y_true,y_pred,k) for k in np.linspace(0, 1, 1000)],axis=0)
pfas = tf.concat([tf.ones((1,)) ,pfas],axis=0)
binSizes = -(pfas[1:]-pfas[:-1])
s = ptas*binSizes
return K.sum(s, axis=0)
# PFA, prob false alert for binary classifier
def binary_PFA(y_true, y_pred, threshold=K.variable(value=0.5)):
y_pred = K.cast(y_pred >= threshold, 'float32')
# N = total number of negative labels
N = K.sum(1 - y_true)
# FP = total number of false alerts, alerts from the negative class labels
FP = K.sum(y_pred - y_pred * y_true)
return FP/N
# P_TA prob true alerts for binary classifier
def binary_PTA(y_true, y_pred, threshold=K.variable(value=0.5)):
y_pred = K.cast(y_pred >= threshold, 'float32')
# P = total number of positive labels
P = K.sum(y_true)
# TP = total number of correct alerts, alerts from the positive class labels
TP = K.sum(y_pred * y_true)
return TP/P
Kenneth Emeka Odoh
Private board: 0.59

Alternatives
● Trees (catboost, lightboost)
Other that could work with careful effort
● HMM
● MARS
● ANFIS
● Logistic regression
● Naive Bayes

Conclusions
● Building models versus building infrastructure trade off.
● Debugging a failing model versus trying out a new model.
● Feature engineering versus better algorithm.
● Neural depth versus training size.
● “Ensembling work when the conditions are right”.
○ Provide empirical justification.
● The place of magic numbers.
● “Leakages may not actually be a bad thing”.
43
Kenneth Emeka Odoh

Thanks for listening
@kenluck2001
44
https://www.linkedin.com/in/kenluck2001/
kenneth.odoh@gmail.com
Kenneth Emeka Odoh

References
Yu, Zhou et al. “Using bidirectional lstm recurrent neural networks to learn high-level abstractions of sequential features for automated scoring of
non-native spontaneous speech.” ASRU (2015).
Ioffe, S. and Szegedy, C. Batch Normalization: Accelerating Deep Network Training by Reducing Internal Covariate Shift. Proceedings of the 32nd
International Conference on Machine Learning, in PMLR 37:448-456 (2015).
Francesco Ricci, Lior Rokach, Bracha Shapira, and Paul B. Kantor. 2010. Recommender Systems Handbook (1st ed.). Springer-Verlag New York, Inc.,
New York, NY, USA.
Goodfellow, I., Bengio, Y., Courville, A. (2016). Deep Learning. MIT Press.
Jeru666, Visualization script, https://www.kaggle.com/jeru666/data-exploration-and-analysis .
Davis, J. & Goadrich, M. (2006). The relationship between Precision-Recall and ROC curves. ICML '06: Proceedings of the 23rd international
conference on Machine learning 233--240, New York, NY, USA
Fei-Fei Li & Justin Johnson & Serena Yeung, (2017), http://cs231n.stanford.edu/slides/2017/cs231n_2017_lecture10.pdf
Srivastava, N., Hinton, G. E., Krizhevsky, A., Sutskever, I. & Salakhutdinov, R. (2014). Dropout: a simple way to prevent neural networks from
overfitting.. Journal of Machine Learning Research, 15, 1929-1958.
45
Kenneth Emeka Odoh

Kaggle kenneth

More Related Content

What's hot (20)

Similar to Kaggle kenneth (20)

Recently uploaded (20)

Kaggle kenneth