xiaohan2012

import tensorflow as tf

checkpoint_file = 'runs/deepwalk/checkpoints/model-25000'

sess = tf.InteractiveSession()

saver = tf.train.import_meta_graph("{}.meta".format(checkpoint_file))
saver.restore(sess, checkpoint_file)

embedding_table = sess.graph.get_operation_by_name('embedding/table')

xiaohan2012

CXX=g++
CPPFLAGS=-std=c++11 -O2 -Wall
LINK_OPTS=-shared -Wl,--export-dynamic
COMPILE_OPTS=-fPIC
INCLUDES=-I /usr/include/python3.5/
LIBS=-L /usr/lib/x86_64-linux-gnu/ -l boost_python-py35

%.so: %.o
        $(CXX) $^ $(CPPFLAGS) $(LINK_OPTS) $(LIBS) -o $@

xiaohan2012

CXX=g++
CPPFLAGS=-std=c++11 -O2 -Wall
ODIR=obj
DEPS=-I boost

%: $(ODIR)/%.o
        $(CXX) -o $@ $^ $(CPPFLAGS)

$(ODIR)/%.o: %.cpp $(DEPS)
        $(CXX) -c -o $@ $< $(CPPFLAGS)

xiaohan2012

CXX=g++
CPPFLAGS=-std=c++11 -O2 -Wall
ODIR=obj
DEPS=-I boost

%: $(ODIR)/%.o
        $(CXX) -o $@ $^ $(CPPFLAGS)

$(ODIR)/%.o: %.cpp $(DEPS)
        $(CXX) -c -o $@ $< $(CPPFLAGS)

xiaohan2012

• figure
• event graph
• timeline

xiaohan2012

• mistake about the bound:
  • charikar's paper
  • 4-approx paper
• synthetic experiment design
  • dynamics of conversations
  • parameters?
    • event/total participant number: influences the event
    • event/total duration: influences the event
  • should we compare the pred tree with actual tree?
• exploration/exploitation experiment

xiaohan2012

PCST:

• why I work on PCST
  • the road map: PCST->MST->Quota->Budget
  • PCST->MST
  • MST = Quota
  • Quota -> Budget: how the 5-approximation comes?
  • that's for undirected case
• incorrectness of the proof on the greedy primal-dual method
• the chapter

xiaohan2012

# -*- coding: utf-8 -*-

"""
Parser for Stanford Dependency Parser output to parse back the original data structure in Python

The parser link:

http://nlp.stanford.edu/software/nndep.shtml

The visualization can be created by generating .dot file

xiaohan2012

"""
Comparing adagrad, adadelta and constant learning in gradient descent(the seddle point function y^2 - x^2)

Reference:
1. comparison on several learning rate update scheme: http://ml.memect.com/archive/2014-12-12/short.html#3786866375172817
2. Saddle point, http://en.wikipedia.org/wiki/Saddle_point

"""
import numpy as np
import theano

xiaohan2012

all: btree types.h rule table cyk main
    g++-4.7 -std=c++11 *.o -o main

main: main.cpp
    g++-4.7 -c -std=c++11 main.cpp

cyk: cyk.cpp
    g++-4.7 -c -std=c++11 cyk.cpp

btree: btree.cpp