forked from zhangjiekui/kaggle-avazu
-
Notifications
You must be signed in to change notification settings - Fork 0
/
README
109 lines (64 loc) · 3.2 KB
/
README
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
4 Idiots' Approach for Click-through Rate Prediction
====================================================
Our team consists of:
Name Kaggle ID Affiliation
====================================================================
Yu-Chin Juan guestwalk National Taiwan University (NTU)
Wei-Sheng Chin mandora National Taiwan University (NTU)
Yong Zhuang yolicat National Taiwan University (NTU)
Michael Jahrer Michael Jahrer Opera Solutions
Our final model is an ensemble of NTU's model and Michael's model. Michael's
model is based on his work in Opera Solutions, so he cannot release his part.
Therefore, in the codes and documents we only present NTU's model.
This README introduces how to run our code up. For the introduction to our
approach, please see
http://www.csie.ntu.edu.tw/~r01922136/slides/kaggle-avazu.pdf
The model we use for this competition is called `field-aware factorization
machines.' We have released a package for this model at:
http://www.csie.ntu.edu.tw/~r01922136/libffm
System Requirement
==================
- 64-bit Unix-like operating system
- Python 3
- g++ (with C++11 and OpenMP support)
- pandas (required if you want to run the `bag' part. See `Step-by-step'
below.)
Step-by-step
============
Our solution is an ensemble of 20 models. It is organized into the following
three parts:
name public score private score description
===========================================================================
base 0.3832 0.3813 2 basic models
bag 0.3826 0.3807 2 models using bag features.
ensemble 0.3817 0.3797 an ensemble of the above 4
models and 16 new small models
Because the `bag' part consumes a huge amount of memory (more than 64GB), and
the `ensemble' part takes a long time to run, this instruction guides you to
run our `base' part first. If you want reproduce our best result, please run the
commands in the final step on a suitable machine.
1. First, please use the following command to run a tiny example up
$ ./run.sh x
2. Create a symbolic link to the training dataset
$ ln -sf <training_set_path> tr.r0.csv
3. Add a dummy label to the test set
$ ./add_dummy_label.py <test_set_path> va.r0.csv
4. Checksum
$ md5sum tr.r0.csv va.r0.csv
f5d49ff28f41dc993b9ecb2372abb033 tr.r0.csv
6edd380a5897bc16b61c5a626062f7b3 va.r0.csv
5. Reproduce our base submission
$ ./run.sh 0
Note: base.r0.prd is the submission file
6. (optional) Reproduce our best submission
$ ./run_all.sh x
If success, then run
$ ./run_all.sh 0
Note: The algorithm in the `bag' part is non-deterministic. That is, the
result can be slightly different when you run it two or more times.
==============
If you want to trace these codes, please be prepared that it will take some
efforts. We do not have enough time to polish the codes here to improve the
readability. Sorry about it.
For any questions and comments, please send your email to:
Yu-Chin ([email protected])