Skip to content

Instantly share code, notes, and snippets.

View dcslin's full-sized avatar

Shicong dcslin

26 followers · 78 following
  • Singapore
  • 00:17 (UTC +08:00)
View GitHub Profile
@dcslin
dcslin / understand-requiment-kaggle-example.md
Last active August 12, 2018 17:06
this is a test
Competition Goal Metric Dataset
Porto Seguro’s Safe Driver Prediction File Claim in 1 year? Normalized Gini Coefficient Relational - Encoded
Home Credit Default Risk Default? AOROC Relational
Titanic: Machine Learning from Disaster Survive? Accuracy Relational
House Prices: Advanced Regression Techniques how much $ RMSE Relational
Digit Recognizer What number Accuracy images
Santander Customer Satisfaction customer happy? ROC Relational - Encoded
Toxic Comment Classification Challenge Multiple labels(6*): toxicity types w/ probs ROC text
@dcslin
dcslin / lstm_predict
Created October 30, 2018 06:21
lstm_predict
python 8_extract_lstm.py
(1000, 2)
(1000,)
2900 candidates has 2900
2901 candidates has 2901
2902 candidates has 2902
2903 candidates has 2903
2904 candidates has 2904
2905 candidates has 2905
2906 candidates has 2906
@dcslin
dcslin / pgm-readings.md
Last active November 30, 2018 12:52
probabilistic graphical model - further readings
@dcslin
dcslin / clustering-news-article.md
Last active December 11, 2018 06:07
news article clustering

cluster a list of news/article and group them if regardings same piece of news.

online resources

  • google news articles

    • no timing feature in the dataset: link
    • clustering stories
    • The source ranking involves many things. Is there original content? The timeliness. Coverage of recent developments? The relevancy to the cluster at hand. In some cases, is there local relevancy? Is there content from a local source with local content? link

  • is a topic modeling problem link

  • modeling link

@dcslin
dcslin / rl.md
Last active December 23, 2018 15:45
reinforcement learning study notes

online resources

source accessable inaccessable
stanford course cs234 slides assignment, solution, lecture
book link avail online solution
udacity link current all content -
edx link - content
berkerly link, current slides, homework lecture
coursera specialization finance link - all
@dcslin
dcslin / cs.vim
Last active January 22, 2019 06:15
vim-cheat-sheet-easy-forget
" helpers for some easily forget features of vim
" reformat by textwidth
set textwidth=80
" then make selection
gq
" colorcolumn
set colorcolum=80
@dcslin
dcslin / sh.sh
Last active January 22, 2019 06:45
singa-helpers
cmake \
-DPYTHON_EXECUTABLE:FILEPATH=/usr/bin/python \
-DPYTHON_INCLUDE_DIR:PATH=/usr/include/python3.2 \
-DPYTHON_LIBRARY:FILEPATH=/usr/lib/libpython3.2.so \
..
# test github ssh key
ssh -T git@github.com
# docker mount host dir to container
@dcslin
dcslin / run.sh
Created February 16, 2019 06:45
compile vim8.1 on ubuntu16.04
# tested on ubuntu 16.04 -- Feb 16 2019
# install ncurses lib (optional, dependency of vim)
wget https://invisible-mirror.net/archives/ncurses/ncurses-6.1.tar.gz
tar xzf ncurses-6.1.tar.gz
cd ncurses-6.1
./configure --prefix=$HOME/ncurses
make
make install
@dcslin
dcslin / dependencies.yaml
Last active March 11, 2019 10:49
dependencies.yaml
package:
name: swig
version: 3.0.12
run:
- libgcc-ng >=7.3.0
- libstdcxx-ng >=7.3.0
- pcre >=8.41,<9.0a0
------------------------------------------------------------------------------------
{% set version = "3.0.10" %}
@dcslin
dcslin / gist:1478c3d2221e7bde9e0c3307e0e76043
Created April 2, 2019 11:03
Epoch 10 [64.86s]: train_loss = 0.235510
Epoch 20 [46.46s]: train_loss = 0.195671
Epoch 30 [62.66s]: train_loss = 0.179149
Took 1558.2378838062286 seconds for training.
Took 2446.4982657432556 seconds for prediction.
MAP: 0.020114
NDCG: 0.039552
Precision@K: 0.004189
Recall@K: 0.111101
1554198866.5532758