-
Notifications
You must be signed in to change notification settings - Fork 0
/
log2015_05_25.html
300 lines (300 loc) · 29.1 KB
/
log2015_05_25.html
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
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
<!DOCTYPE html>
<html lang='en'>
<head>
<title>
pgmpy Log of 25-05-2015
</title>
<link rel='stylesheet' type='text/css' href='css/bootstrap.min.css' media='screen' />
</head>
<body>
<div class='container'>
<h2>Log of the <code>#pgmpy</code> IRC Channel</h2><br/>
<h3>All the times shown here presently are in Indian Standard Time(IST) +0530Hrs<h3/>
<h3>Date : 25-05-2015 </h3><br/><br/>
<br/>
<kbd>[16:28:13]</kbd> <strong>[connected at Mon May 25 16:28:13 2015]</strong><br/>
<kbd>[16:28:20]</kbd> <strong>[I have joined #pgmpy]</strong><br/>
<kbd>[16:28:25]</kbd> <code><ankurankan></code> kislay, Hi<br/>
<kbd>[16:28:27]</kbd> <code><kislay></code> do we have a meeting?<br/>
<kbd>[16:28:31]</kbd> <code><ankurankan></code> Yes<br/>
<kbd>[16:28:36]</kbd> <code><kislay></code> okay!<br/>
<kbd>[16:28:46]</kbd> <code><kislay></code> from 10, right?<br/>
<kbd>[16:29:05]</kbd> <code><ankurankan></code> Yes<br/>
<kbd>[16:29:13]</kbd> <code><kislay></code> cool<br/>
<kbd>[16:33:20]</kbd> <code><ankurankan></code> Hi everyone<br/>
<kbd>[16:33:33]</kbd> <code><vivekjain></code> Hi<br/>
<kbd>[16:33:46]</kbd> <code><palashahuja></code> hello<br/>
<kbd>[16:33:50]</kbd> <code><ankurankan></code> How was the community bonding period ? Everyone familiar with code ?<br/>
<kbd>[16:33:51]</kbd> <code><kislay></code> Hi everyone<br/>
<kbd>[16:33:51]</kbd> <code><pratyaksh></code> hello<br/>
<kbd>[16:34:49]</kbd> <code><kislay></code> I have figured out the input and output variables along with the fine tuning variables that opengm uses to solve a LP_relaxation. I have some doubts and really need to discuss the whole framework and the practices that are to be followed with pratyaksh, Jaidev and abinash. I have pretty much summarised everything I have done via some blog feeds.<br/>
<kbd>[16:35:07]</kbd> <code><kislay></code> hmm..for the code to be written,<br/>
<kbd>[16:35:30]</kbd> <code><kislay></code> I think I need one meeting with abinash<br/>
<kbd>[16:35:37]</kbd> <code><kislay></code> and Jaidev together<br/>
<kbd>[16:35:48]</kbd> <code><pratyaksh_></code> Even I am yet to discuss the plan with them<br/>
<kbd>[16:35:51]</kbd> <code><jaidev></code> hey guys<br/>
<kbd>[16:35:55]</kbd> <code><kislay></code> hi jaidev<br/>
<kbd>[16:36:04]</kbd> <code><jaidev></code> Hi abhijeet, did you see me mail?<br/>
<kbd>[16:36:08]</kbd> <code><jaidev></code> *my<br/>
<kbd>[16:36:24]</kbd> <code><kislay></code> yup, just saw. Will improve upon the points you mentioned.<br/>
<kbd>[16:36:36]</kbd> <code><ankurankan></code> You guys should send emails on the mailing list whenever you have any problem<br/>
<kbd>[16:37:00]</kbd> <code><ankurankan></code> That way somebody else can reply if your mentor is not available<br/>
<kbd>[16:37:22]</kbd> <code><kislay></code> okay. Will do. jaidev, we need to have a meet to discuss few important things...<br/>
<kbd>[16:37:27]</kbd> <code><pratyaksh_></code> fine<br/>
<kbd>[16:37:34]</kbd> <code><kislay></code> along with abinash<br/>
<kbd>[16:37:58]</kbd> <code><jaidev></code> pratyaksh_: I also got your google doc.<br/>
<kbd>[16:38:06]</kbd> <code><jaidev></code> It doesn't seem like it is finished<br/>
<kbd>[16:38:13]</kbd> <code><jaidev></code> Is there more you are going to write?<br/>
<kbd>[16:38:22]</kbd> <code><pratyaksh_></code> right, it is definitely not done<br/>
<kbd>[16:38:37]</kbd> <code><jaidev></code> Alright, what's your plan for the week?<br/>
<kbd>[16:38:46]</kbd> <code><ankurankan></code> pratyaksh_, Could you please share the doc with all the mentors ?<br/>
<kbd>[16:38:47]</kbd> <code><jaidev></code> lol<br/>
<kbd>[16:38:48]</kbd> <code><pratyaksh_></code> i need to read up a bit and then i'll complete it<br/>
<kbd>[16:39:02]</kbd> <code><pratyaksh_></code> sure<br/>
<kbd>[16:39:07]</kbd> <code><jaidev></code> Okay, ping me when you are done.<br/>
<kbd>[16:39:36]</kbd> <code><pratyaksh_></code> btw, as per the timeline, i can start coding what i planned for the first week<br/>
<kbd>[16:39:51]</kbd> <code><jaidev></code> What is the plan for the first week?<br/>
<kbd>[16:39:58]</kbd> <code><jaidev></code> kislay: Send me the link to those slides.<br/>
<kbd>[16:40:33]</kbd> <code><pratyaksh_></code> mostly implement forward, rejection and importance sampling<br/>
<kbd>[16:41:01]</kbd> <code><kislay></code> jaidev, the whole slides you mean. Okay. I put them out from some of the Sontag video lectures and online classes.<br/>
<kbd>[16:41:21]</kbd> <code><pratyaksh_></code> ^which wil be a useful utility to have, and will be useful in the later stages too<br/>
<kbd>[16:42:44]</kbd> <code><jaidev></code> ankurankan: Can you fix the stuff here - https://github.com/pgmpy/pgmpy_notebook/blob/master/Probabilistic%20Graphical%20Models%20using%20pgmpy.ipynb<br/>
<kbd>[16:42:47]</kbd> <code><jaidev></code> ><br/>
<kbd>[16:42:47]</kbd> <code><jaidev></code> ?<br/>
<kbd>[16:43:17]</kbd> <code><jaidev></code> It looks like it is from the days when Python2 was being supported. I tried a lot to fix it, but most of what I did was guesswork.<br/>
<kbd>[16:43:37]</kbd> <code><jaidev></code> If those notebooks are fixed, I can get acquainted with the codebase much faster.<br/>
<kbd>[16:44:10]</kbd> <code><ankurankan></code> jaidev, I will try to fix it ASAP<br/>
<kbd>[16:44:16]</kbd> <code><kislay></code> hi abinash<br/>
<kbd>[16:44:21]</kbd> <code><abinash></code> Hi all<br/>
<kbd>[16:44:25]</kbd> <code><abinash></code> hi kislay<br/>
<kbd>[16:44:27]</kbd> <code><jaidev></code> ankurankan: Thanks.<br/>
<kbd>[16:44:32]</kbd> <code><jaidev></code> hey abinash<br/>
<kbd>[16:44:36]</kbd> <code><abinash></code> hi jaidev<br/>
<kbd>[16:46:22]</kbd> <code><ankurankan></code> palashahuja, What are your plans for the next week?<br/>
<kbd>[16:48:47]</kbd> <code><abinash></code> kislay or pratyaksh your plans?<br/>
<kbd>[16:48:56]</kbd> <code><jaidev></code> is Vivek here?<br/>
<kbd>[16:49:02]</kbd> <code><vivekjain></code> jaidev: Hi<br/>
<kbd>[16:49:20]</kbd> <code><jaidev></code> Hey<br/>
<kbd>[16:50:13]</kbd> <code><jaidev></code> vivekjain: Your document is complete or is there anything left to add<br/>
<kbd>[16:50:13]</kbd> <code><pratyaksh></code> i'll stick to my proposal timeline. will implement forward sampling and rejection sampling<br/>
<kbd>[16:50:13]</kbd> <code><jaidev></code> ?<br/>
<kbd>[16:50:49]</kbd> <code><abinash></code> pratyaksh: so are you now familiar with the code base?<br/>
<kbd>[16:50:58]</kbd> <code><vivekjain></code> jaidev: I will add things in the document as i progress through the project implementation.<br/>
<kbd>[16:51:02]</kbd> <code><abinash></code> the same question for everyone else?<br/>
<kbd>[16:51:35]</kbd> <code><pratyaksh></code> i will only deal with the inference module. so yes i'm familiar with that<br/>
<kbd>[16:51:57]</kbd> <code><abinash></code> Cool<br/>
<kbd>[16:52:08]</kbd> <code><kislay></code> Sorry... My net broke :(<br/>
<kbd>[16:52:08]</kbd> <code><ankurankan></code> pratyaksh, Could you please share the API with us before you start coding it?<br/>
<kbd>[16:52:26]</kbd> <code><pratyaksh></code> fine I'll do that<br/>
<kbd>[16:53:09]</kbd> <code><vivekjain></code> abinash: Yes i am familiar with the readwrite module .<br/>
<kbd>[16:54:08]</kbd> <code><abinash></code> pratyaksh: The API should be more of less simlar to the ones present in exact inference ones.<br/>
<kbd>[16:54:13]</kbd> <code><jaidev></code> vivekjain: But your plan for this week, as in the document, is finalized, right?<br/>
<kbd>[16:54:23]</kbd> <code><pratyaksh></code> abinash: yes i'm keeping that in mind<br/>
<kbd>[16:54:28]</kbd> <code><ankurankan></code> vivekjain, Could you work on XMLBIF before you start with ProbModelXML?<br/>
<kbd>[16:54:28]</kbd> <code><vivekjain></code> jaidev: For this week, i will list down the TODO's for the ProbModelXML file format and implement these two TODOs over the period of two weeks<br/>
<kbd>[16:54:40]</kbd> <code><abinash></code> kislay: Are you familar with the code base?<br/>
<kbd>[16:55:07]</kbd> <code><kislay></code> abinash: , I have been running opengm for relaxations. I think I have a basic idea about its use. I was hoping to get ahead with the discussion on how to get it done for pgmpy<br/>
<kbd>[16:55:40]</kbd> <code><ankurankan></code> vivekjain, I mean switching week 4 with week 1?<br/>
<kbd>[16:55:54]</kbd> <code><vivekjain></code> ankurankan: ok.<br/>
<kbd>[16:55:58]</kbd> <code><abinash></code> kislay: Okay. Lets discuss after this meeting.<br/>
<kbd>[16:56:12]</kbd> <code><abinash></code> jaidev: It would be fine for you. Right?<br/>
<kbd>[16:56:47]</kbd> <code><jaidev></code> abinash: Sorry what would be fine?<br/>
<kbd>[16:57:07]</kbd> <code><vivekjain></code> jaidev: ankurankan I will update the document and plan for the same.<br/>
<kbd>[16:57:09]</kbd> <code><kislay></code> jaidev: a small chat, i hope<br/>
<kbd>[16:57:16]</kbd> <code><ankurankan></code> vivekjain, Cool<br/>
<kbd>[16:57:17]</kbd> <code><abinash></code> jaidev: A small meeting with kislay regarding his plans.<br/>
<kbd>[16:57:25]</kbd> <code><jaidev></code> Oh yes, sure<br/>
<kbd>[16:57:50]</kbd> <code><jaidev></code> vivekjain: all right<br/>
<kbd>[16:58:22]</kbd> <code><jaidev></code> So who's coming to PyCon?<br/>
<kbd>[16:59:44]</kbd> <code><ankurankan></code> jaidev, I might be coming<br/>
<kbd>[17:01:36]</kbd> <code><jaidev></code> Propose something quickly then :)<br/>
<kbd>[17:03:08]</kbd> <code><ankurankan></code> jaidev, I am on a really tight schedule right now.. We still have 5-6 days left ?<br/>
<kbd>[17:04:12]</kbd> <code><ankurankan></code> Before the deadline?<br/>
<kbd>[17:05:34]</kbd> <code><ankurankan></code> pratyaksh, You mentioned some implementation related doc that you have.. Could you share it me?<br/>
<kbd>[17:05:58]</kbd> <code><jaidev></code> ankurankan: 31st is the deadline, so just do it before that.<br/>
<kbd>[17:06:37]</kbd> <code><ankurankan></code> jaidev, Yeah I will.. Will also try to write a proposal for the workshop..<br/>
<kbd>[17:06:48]</kbd> <code><jaidev></code> Cool<br/>
<kbd>[17:06:53]</kbd> <code><ankurankan></code> vivekjain, You also.. Please share the doc<br/>
<kbd>[17:07:20]</kbd> <code><vivekjain></code> ankurankan: i have already shared my doc .<br/>
<kbd>[17:07:37]</kbd> <code><pratyaksh></code> ankur: i'll be modifying it.. but have shared it for now<br/>
<kbd>[17:07:51]</kbd> <code><ankurankan></code> vivekjain, Didn't get a mail.. Can you give me the link?<br/>
<kbd>[17:08:30]</kbd> <code><vivekjain></code> ankurankan: https://docs.google.com/document/d/19u21r7Ud-mkzbomvsCpI957K6FSBq0I_S6jvTf1ACak/edit This is the link<br/>
<kbd>[17:11:25]</kbd> <code><ankurankan></code> So, anything else for today?<br/>
<kbd>[17:14:35]</kbd> <code><jaidev></code> kislay: abinash I'm need to finish off something for about 20 mins or so. Let's have a hangout at ~ 11.10 pm?<br/>
<kbd>[17:14:49]</kbd> <code><kislay></code> jaidev: cool<br/>
<kbd>[17:15:10]</kbd> <code><jaidev></code> abinash: suits you?<br/>
<kbd>[17:15:28]</kbd> <code><pratyaksh></code> we should share a doc where everyone can write some points that they'd like to discuss in the weekly meeting. it'll be more efficient i guess.<br/>
<kbd>[17:16:07]</kbd> <code><pratyaksh></code> and, can we move to hangouts? :p<br/>
<kbd>[17:18:08]</kbd> <code><ankurankan></code> pratyaksh, I agree. I have just created a folder on drive and shared with you all.. Please add all your GSoC related docs there so that everybody can see..<br/>
<kbd>[17:18:28]</kbd> <code><ankurankan></code> Also I will create a file for having points for discussion in the next meeting<br/>
<kbd>[17:19:01]</kbd> <code><pratyaksh></code> sounds great<br/>
<kbd>[17:19:12]</kbd> <code><kislay></code> Cool<br/>
<kbd>[17:21:41]</kbd> <code><jaidev></code> okay guys, see you soon, good night<br/>
<kbd>[17:21:43]</kbd> <code><ankurankan></code> And also please create docs with descriptive names (not something like implementation) .. Maybe add your project name before that<br/>
<kbd>[17:22:38]</kbd> <code><ankurankan></code> Anything else for today?<br/>
<kbd>[17:22:47]</kbd> <code><palashahuja></code> ankurankan: please make the logs available<br/>
<kbd>[17:23:02]</kbd> <code><kislay></code> Logs are working .<br/>
<kbd>[17:23:25]</kbd> <code><ankurankan></code> palashahuja, They are working now<br/>
<kbd>[17:25:40]</kbd> <code><ankurankan></code> Please make sure that all of you add all your docs to that folder today only.<br/>
<kbd>[17:25:51]</kbd> <code><ankurankan></code> I think that's it for today.<br/>
<kbd>[17:25:59]</kbd> <code><ankurankan></code> Bye..<br/>
<kbd>[17:33:00]</kbd> <code><kislay></code> hey abinash<br/>
<kbd>[17:33:08]</kbd> <code><abinash></code> hey kislay<br/>
<kbd>[17:55:35]</kbd> <code><abinash></code> hey kislay and jaidevd<br/>
<kbd>[17:55:39]</kbd> <code><jaidevd></code> hey<br/>
<kbd>[17:55:49]</kbd> <code><abinash></code> kislay: are you there?<br/>
<kbd>[17:55:50]</kbd> <code><kislay></code> hi<br/>
<kbd>[17:55:52]</kbd> <code><kislay></code> yes<br/>
<kbd>[17:55:53]</kbd> <code><abinash></code> Cool<br/>
<kbd>[17:56:03]</kbd> <code><abinash></code> We can start the discussion. kislay go ahead.<br/>
<kbd>[17:56:23]</kbd> <code><kislay></code> First we need to get an input of factor graphs as the representation of the model<br/>
<kbd>[17:56:42]</kbd> <code><abinash></code> Okay.<br/>
<kbd>[17:56:54]</kbd> <code><abinash></code> FactorGraph model is implemented. Right?<br/>
<kbd>[17:56:56]</kbd> <code><kislay></code> means there should be a way we can get pgmpy understand our f.g model<br/>
<kbd>[17:57:03]</kbd> <code><kislay></code> I asked ankurankan<br/>
<kbd>[17:57:09]</kbd> <code><kislay></code> he replied negative<br/>
<kbd>[17:57:23]</kbd> <code><kislay></code> we don't have it yet. AFAIK<br/>
<kbd>[17:57:31]</kbd> <code><abinash></code> Ahh...you asked whether we have a reader to convert from UAI file format to FactorGraph<br/>
<kbd>[17:57:42]</kbd> <code><abinash></code> that is not there, but you can manually create one. Right?<br/>
<kbd>[17:57:58]</kbd> <code><kislay></code> hmm.. oh<br/>
<kbd>[17:58:17]</kbd> <code><kislay></code> facor graphs are a bit different. I think yes<br/>
<kbd>[17:58:32]</kbd> <code><abinash></code> BTW last year I implemented UAI file format reader in Shogun. You can use the python bindings, but it won't be worth the time.<br/>
<kbd>[17:58:35]</kbd> <code><kislay></code> I will have to see...what do you think jaidevd<br/>
<kbd>[17:58:45]</kbd> <code><kislay></code> oh cool abinash<br/>
<kbd>[17:59:09]</kbd> <code><abinash></code> What is the major issue that you are facing if there is no UAI file reader.<br/>
<kbd>[17:59:11]</kbd> <code><abinash></code> ?<br/>
<kbd>[17:59:24]</kbd> <code><abinash></code> Are you blocked somewhere?<br/>
<kbd>[18:00:03]</kbd> <code><kislay></code> yes, that is a preliminary issue. ankurankan told me that he will take care of that thing . Instead I should carry on with the main point<br/>
<kbd>[18:00:15]</kbd> <code><jaidevd></code> Forget shogun for now, it shouldn't be a pgmpy dependency<br/>
<kbd>[18:00:27]</kbd> <code><abinash></code> Yeah. I totally agree.<br/>
<kbd>[18:00:46]</kbd> <code><abinash></code> See instead of using a file, why don't you create a graph manually?<br/>
<kbd>[18:00:53]</kbd> <code><kislay></code> next comes the fact that, we write inference algorithms.<br/>
<kbd>[18:00:53]</kbd> <code><kislay></code> For solving LP_relaxations, we need a Linear programming solver... we have CPLEX in OpenGM for that.<br/>
<kbd>[18:00:59]</kbd> <code><kislay></code> oh...<br/>
<kbd>[18:01:01]</kbd> <code><abinash></code> It would be only a one time effort (although a tedious one:()<br/>
<kbd>[18:01:02]</kbd> <code><kislay></code> yes abinash<br/>
<kbd>[18:01:06]</kbd> <code><jaidevd></code> What about cvxopt?<br/>
<kbd>[18:01:24]</kbd> <code><abinash></code> Yeah. I would suggest cvxopt also.<br/>
<kbd>[18:01:28]</kbd> <code><kislay></code> that is totally fine. I was not sure about it. But that will be much better option for now.<br/>
<kbd>[18:01:51]</kbd> <code><ankurankan></code> kislay, Can you point me to some link where I can find some factor graph example in UAI format?<br/>
<kbd>[18:02:09]</kbd> <code><kislay></code> cool . I will note cvxopt.<br/>
<kbd>[18:02:12]</kbd> <code><abinash></code> http://graphmod.ics.uci.edu/uai08/FileFormat<br/>
<kbd>[18:02:18]</kbd> <code><kislay></code> ankurankan: yup .sure<br/>
<kbd>[18:03:27]</kbd> <code><abinash></code> Furthermore, I don't think CPLEX is free of cost.<br/>
<kbd>[18:03:36]</kbd> <code><jaidevd></code> I'm not saying we should definitely move to cvxopt, just that it is much more popular and users might be more comfortable with it. So let's try to use it unless we have a very strong reason not to.<br/>
<kbd>[18:04:07]</kbd> <code><abinash></code> cvxopt provides us with the same kind of flexibility as CPLEX.<br/>
<kbd>[18:04:19]</kbd> <code><kislay></code> that would be great abinash<br/>
<kbd>[18:04:31]</kbd> <code><kislay></code> yup. agree jaidevd<br/>
<kbd>[18:04:34]</kbd> <code><abinash></code> But it would be good to have a benchmark before making any final decision. What are your suggestions jaidevd and kislay ?<br/>
<kbd>[18:04:58]</kbd> <code><kislay></code> again that is a cool thing. But do we have other options?<br/>
<kbd>[18:05:06]</kbd> <code><jaidevd></code> What sort of benchmark? As in, compare the performance of cvxopt versus cplex?<br/>
<kbd>[18:05:42]</kbd> <code><abinash></code> See for linear programming solvers we have a lot of options. Why don't you do some benchmarking over a bunch of libraries before finalizing?<br/>
<kbd>[18:05:53]</kbd> <code><abinash></code> It would be a great content for blog as well :)<br/>
<kbd>[18:05:55]</kbd> <code><kislay></code> yeah...we need to have atleat two python workable options<br/>
<kbd>[18:06:10]</kbd> <code><kislay></code> okay. Noted.<br/>
<kbd>[18:06:11]</kbd> <code><jaidevd></code> There should be many<br/>
<kbd>[18:06:26]</kbd> <code><jaidevd></code> But don't spend too much time on this. Just enough to make a decision<br/>
<kbd>[18:06:35]</kbd> <code><kislay></code> yup. So first comes the benchmarking.<br/>
<kbd>[18:06:36]</kbd> <code><abinash></code> Yeah. I agree.<br/>
<kbd>[18:06:39]</kbd> <code><kislay></code> okay. I get it<br/>
<kbd>[18:06:41]</kbd> <code><abinash></code> Don't spend a lot of time.<br/>
<kbd>[18:06:49]</kbd> <code><jaidevd></code> Let's get the features in first, we can scale later.<br/>
<kbd>[18:06:50]</kbd> <code><kislay></code> yup<br/>
<kbd>[18:06:55]</kbd> <code><abinash></code> Try to finish it within a day or two.<br/>
<kbd>[18:06:59]</kbd> <code><kislay></code> so the 3rd and last part..<br/>
<kbd>[18:07:01]</kbd> <code><abinash></code> Not more than that.<br/>
<kbd>[18:07:18]</kbd> <code><kislay></code> Then comes the inference part where I will have to optimise over the cuts of the marginal polytopes<br/>
<kbd>[18:07:28]</kbd> <code><abinash></code> Okay.<br/>
<kbd>[18:07:31]</kbd> <code><kislay></code> Now I have gone through humongous amount of theory<br/>
<kbd>[18:07:38]</kbd> <code><kislay></code> related to the ^^^<br/>
<kbd>[18:07:52]</kbd> <code><kislay></code> such is that it become a glob of mass in my head<br/>
<kbd>[18:08:11]</kbd> <code><kislay></code> I have a cool SOntag's MPLP code to start with<br/>
<kbd>[18:08:41]</kbd> <code><kislay></code> abinash: I got it from the paper you marked beside the project (it is basically a matlab prototype )<br/>
<kbd>[18:09:08]</kbd> <code><abinash></code> Thats good. Why don't you start with that? Try to play around with it and try to understand how it is implemented?<br/>
<kbd>[18:09:27]</kbd> <code><abinash></code> Don't just try to imitate the MATLAB code in python, because it won't work :P<br/>
<kbd>[18:09:41]</kbd> <code><kislay></code> Except that, i think i will be needing a lot of inspiration from you guys on how to get it implemented in python<br/>
<kbd>[18:09:49]</kbd> <code><jaidevd></code> hahahaha<br/>
<kbd>[18:09:53]</kbd> <code><abinash></code> Could you just point to the code?<br/>
<kbd>[18:09:55]</kbd> <code><kislay></code> yeah abinash , that is what i need to be sure of.<br/>
<kbd>[18:10:12]</kbd> <code><jaidevd></code> Put up the code in a temporary repo<br/>
<kbd>[18:10:24]</kbd> <code><abinash></code> See in the first iteration don't think about Cythonizing it or doing any fancy stuffs.<br/>
<kbd>[18:10:33]</kbd> <code><jaidevd></code> We'll take a look and try to see how to best translae that<br/>
<kbd>[18:10:34]</kbd> <code><abinash></code> Implement using simple python<br/>
<kbd>[18:11:05]</kbd> <code><abinash></code> and after that we would be moving incrementally, optimizing the slowest part and moving on.<br/>
<kbd>[18:11:09]</kbd> <code><abinash></code> Your suggestions jaidevd ?<br/>
<kbd>[18:11:15]</kbd> <code><kislay></code> okay. I will do both and agree both abinash , jaidevd<br/>
<kbd>[18:11:22]</kbd> <code><kislay></code> will do both<br/>
<kbd>[18:11:31]</kbd> <code><jaidevd></code> BTW if you're looking for examples of massive amounts of MATLAB translated code - look at github.com/scikit-signal/pytftb<br/>
<kbd>[18:11:41]</kbd> <code><kislay></code> woah...cool<br/>
<kbd>[18:11:44]</kbd> <code><kislay></code> thanks jaidevd<br/>
<kbd>[18:12:09]</kbd> <code><jaidevd></code> In any case, I think I'll first have to look at that code to see how "usable" it is in Python.<br/>
<kbd>[18:12:10]</kbd> <code><abinash></code> haha... jaidevd so you had really spent some time in converting MATLAB shit to python code?<br/>
<kbd>[18:12:17]</kbd> <code><kislay></code> haha<br/>
<kbd>[18:12:23]</kbd> <code><jaidevd></code> Are both of you convinced that this is the correct starting point?<br/>
<kbd>[18:12:24]</kbd> <code><abinash></code> wondering if anybody is brave enough to do this :P<br/>
<kbd>[18:12:31]</kbd> <code><kislay></code> Matlab is shit, eh? abinash ;)<br/>
<kbd>[18:12:57]</kbd> <code><jaidevd></code> abinash: I do it for three hours in the morning everyday.<br/>
<kbd>[18:13:14]</kbd> <code><kislay></code> jaidevd: I need suggestion. THis is completely what i think i got to understand from analysing OPenGM.<br/>
<kbd>[18:13:22]</kbd> <code><abinash></code> jaidevd: Haha... :D<br/>
<kbd>[18:13:49]</kbd> <code><kislay></code> I broke into 3 points that I presented here<br/>
<kbd>[18:14:05]</kbd> <code><kislay></code> please suggest.<br/>
<kbd>[18:14:28]</kbd> <code><kislay></code> if you see me getting in something horrible here.<br/>
<kbd>[18:14:32]</kbd> <code><kislay></code> into*<br/>
<kbd>[18:15:14]</kbd> <code><abinash></code> Could you again write those three steps. Have to scroll a lot :P<br/>
<kbd>[18:15:30]</kbd> <code><kislay></code> First we need to get an input of factor graphs insie the model<br/>
<kbd>[18:15:30]</kbd> <code><kislay></code> next comes the fact that, we write inference algorithms.<br/>
<kbd>[18:15:30]</kbd> <code><kislay></code> For solving LP_relaxations, we need a Linear programming solver... we have CPLEX in OpenGM for that.<br/>
<kbd>[18:15:30]</kbd> <code><kislay></code> Then comes the inference part where I will have to optimise over the cuts of the marginal polytopes<br/>
<kbd>[18:15:35]</kbd> <code><jaidevd></code> I'll have to take a look at OpenGM, never even heard of it until today<br/>
<kbd>[18:15:50]</kbd> <code><kislay></code> oh jaidevd , it is a cool thing<br/>
<kbd>[18:16:07]</kbd> <code><kislay></code> jaidevd: go for the pdf manual. That thing is awesome.<br/>
<kbd>[18:16:18]</kbd> <code><abinash></code> So for the first step, you don't need to worry about UAI file reader. Go ahead and create a sample factor graph manually.<br/>
<kbd>[18:16:38]</kbd> <code><abinash></code> For the second one, do the benchmarking of some LP solvers (don't spend too much time)<br/>
<kbd>[18:16:45]</kbd> <code><jaidevd></code> I think you should get started with the input stage. You have to start somewhere, after all. As we go along we'll most probably know a lot more.<br/>
<kbd>[18:17:10]</kbd> <code><abinash></code> for the third one, start with conversion of MATLAB code into simple Python code. jaidevd and I would help you in that<br/>
<kbd>[18:17:38]</kbd> <code><abinash></code> once we have the inference code in place, we would start optimizing (or cythonizing) it.<br/>
<kbd>[18:18:03]</kbd> <code><kislay></code> Let me first get a factor graph inside the pgmpy manually and see how it looks. I think i missed it big time!!!!<br/>
<kbd>[18:18:36]</kbd> <code><jaidevd></code> Yep<br/>
<kbd>[18:18:52]</kbd> <code><kislay></code> hmm..then I will follow you abinash.<br/>
<kbd>[18:19:21]</kbd> <code><kislay></code> btw,I just saw a pdf which someone shared in the folder ankurankan gave us<br/>
<kbd>[18:19:49]</kbd> <code><abinash></code> which pdf?<br/>
<kbd>[18:20:17]</kbd> <code><kislay></code> https://drive.google.com/a/itbhu.ac.in/folderview?id=0B76dzUSFG5DafmxheGxVMTlGWUZJaGJfc2Fadl84d2x3MkpHVW1Xa25IWlNjbi1oSFRwenM&usp=sharing_eid<br/>
<kbd>[18:20:39]</kbd> <code><kislay></code> the sampling algo one<br/>
<kbd>[18:20:51]</kbd> <code><kislay></code> the one with the square brakets<br/>
<kbd>[18:21:50]</kbd> <code><kislay></code> It has structured things in a nice way<br/>
<kbd>[18:22:01]</kbd> <code><kislay></code> do you agree with that structure?? jaidevd abinash<br/>
<kbd>[18:22:27]</kbd> <code><abinash></code> I am not able to see it.<br/>
<kbd>[18:22:45]</kbd> <code><jaidevd></code> let me check<br/>
<kbd>[18:23:17]</kbd> <code><abinash></code> Okay. Now I am able to see it.<br/>
<kbd>[18:23:28]</kbd> <code><ankurankan></code> abinash, Just gave you access<br/>
<kbd>[18:23:30]</kbd> <code><jaidevd></code> Are you referring to pratyaksh's document?<br/>
<kbd>[18:24:30]</kbd> <code><kislay></code> I don't know whose is that<br/>
<kbd>[18:24:41]</kbd> <code><kislay></code> it has a chart<br/>
<kbd>[18:24:46]</kbd> <code><kislay></code> of where to put the code<br/>
<kbd>[18:24:54]</kbd> <code><abinash></code> ankurankan: Is it pratyaksh's doc?<br/>
<kbd>[18:25:13]</kbd> <code><kislay></code> yeah, he is the one working alongside me on approx. algos<br/>
<kbd>[18:25:18]</kbd> <code><kislay></code> so he must be it<br/>
<kbd>[18:25:42]</kbd> <code><abinash></code> Okay.<br/>
<kbd>[18:25:50]</kbd> <code><abinash></code> Anything else kislay and jaidevd ?<br/>
<kbd>[18:26:30]</kbd> <code><kislay></code> nopes. Its just that I will be reporting things on my blogs<br/>
<kbd>[18:26:37]</kbd> <code><kislay></code> I am kinda enjoying them<br/>
<kbd>[18:26:45]</kbd> <code><kislay></code> they are linked with my docs<br/>
<kbd>[18:26:48]</kbd> <code><jaidevd></code> Nothing from me right now.<br/>
<kbd>[18:26:53]</kbd> <code><kislay></code> which are in the shared folder<br/>
<kbd>[18:27:16]</kbd> <code><kislay></code> okay, fine<br/>
<kbd>[18:27:24]</kbd> <code><kislay></code> that's all I had.<br/>
<kbd>[18:27:55]</kbd> <code><abinash></code> Okay. Bye jaidevd and kislay<br/>
<kbd>[18:28:14]</kbd> <code><ankurankan></code> kislay, I am doing a quick implementation of UAI format right now..<br/>
<kbd>[18:28:15]</kbd> <code><jaidevd></code> bye<br/>
<kbd>[18:28:30]</kbd> <code><ankurankan></code> Should be able to give it to you by tomorrow morning<br/>
<kbd>[18:28:45]</kbd> <code><kislay></code> wow. ankurankan thats super fast!<br/>
<kbd>[18:29:37]</kbd> <code><abinash></code> ankurankan: He is super man :P<br/>
<kbd>[18:29:57]</kbd> <code><abinash></code> *ankurankan is super man<br/>
<kbd>[18:31:38]</kbd> <code><kislay></code> okay. Bye guys<br/>
<kbd>[04:58:16]</kbd> <strong>[disconnected at Tue May 26 04:58:16 2015]</strong><br/>
</div>
<script src='https://ajax.googleapis.com/ajax/libs/jquery/1.11.0/jquery.min.js'></script>
<script src='//netdna.bootstrapcdn.com/bootstrap/3.1.1/js/bootstrap.min.js'>
</script>
</body>
</html><br/>