forked from kyleabeauchamp/MSMBuilderTalk
-
Notifications
You must be signed in to change notification settings - Fork 0
/
index.html
797 lines (610 loc) · 20.6 KB
/
index.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
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
<!--
slidedeck: A modification of the Google IO 2012 HTML5 slide template
URL: https://github.com/rmcgibbo/slidedeck
Based on https://github.com/francescolaffi/elastic-google-io-slides, and
ultimately:
Google IO 2012 HTML5 Slide Template
Authors: Eric Bidelman <[email protected]>
Luke Mahe <[email protected]>
URL: https://code.google.com/p/io-2012-slides
-->
<!DOCTYPE html>
<html>
<head>
<title>MSMBuilder3</title>
<meta charset="utf-8">
<meta http-equiv="X-UA-Compatible" content="chrome=1">
<!--<meta name="viewport" content="width=device-width, initial-scale=1.0, minimum-scale=1.0">-->
<!--<meta name="viewport" content="width=device-width, initial-scale=1.0">-->
<!--This one seems to work all the time, but really small on ipad-->
<!--<meta name="viewport" content="initial-scale=0.4">-->
<meta name="apple-mobile-web-app-capable" content="yes">
<link rel="stylesheet" media="all" href="theme/css/default.css">
<link rel="stylesheet" media="all" href="theme/css/custom.css">
<link rel="stylesheet" media="only screen and (max-device-width: 480px)" href="theme/css/phone.css">
<base target="_blank"> <!-- This amazingness opens all links in a new tab. -->
<script data-main="js/slides", src="https://cdnjs.cloudflare.com/ajax/libs/require.js/2.1.14/require.min.js"></script>
<!-- MathJax support -->
<script type="text/x-mathjax-config">
MathJax.Hub.Config({
extensions: ["tex2jax.js"],
jax: ["input/TeX", "output/HTML-CSS"],
showProcessingMessages: false,
tex2jax: {
inlineMath: [ ['$','$'], ["\\(","\\)"] ],
displayMath: [ ['$$','$$'], ["\\[","\\]"] ],
processEscapes: true
},
TeX: {
extensions: ["color.js"]
},
"HTML-CSS": { availableFonts: ["TeX"] }
});
</script>
<script type="text/javascript" src="https://cdn.mathjax.org/mathjax/latest/MathJax.js?config=TeX-AMS-MML_HTMLorMML"></script>
<div style="display:hidden">
\[
\definecolor{data}{RGB}{18,110,213}
\definecolor{unknown}{RGB}{217,86,16}
\definecolor{learned}{RGB}{175,114,176}
\]
</div>
</head>
<body style="opacity: 0">
<slides class="layout-widescreen">
<slide class="title-slide segue nobackground">
<hgroup class="auto-fadein">
<h1> <div style="text-align: center; margin-left: 0em; margin-bottom: -.3em;"> <img height=300 src="http://msmbuilder.org/3.5.0/_static/logo.png"></div></h1>
<h2></h2>
<p> Carlos X. Hernández, Matthew P. Harrigan, M. Muneeb Sultan, Brooke E. Husic<br/> Updated: Jun. 23, 2016 (msmbuilder v3.5)</p>
</hgroup>
</slide>
<slide >
<hgroup>
<h2>Old-School Analysis of MD Data</h2>
<h3></h3>
</hgroup>
<article ><ul>
<li>Analysis happens in walled gardens (Gromacs, Amber, VMD)</li>
<li>Exclusively command line interfaces, C and Fortran code</li>
<li>Duplication of statistical algorithms by non-experts (e.g. chemists, biologists)</li>
<li>Possible code maintainability issues?</li>
</ul></article>
</slide>
<slide >
<hgroup>
<h2>Jarvis Patrick Clustering in Gromacs</h2>
<h3><a href="https://github.com/gromacs/gromacs/blob/master/src/gromacs/gmxana/gmx_cluster.cpp#L502">real code in gromacs</a></h3>
</hgroup>
<article ><pre class="prettyprint" data-lang="c++">
static void jarvis_patrick(int n1, real **mat, int M, int P,
real rmsdcut, t_clusters *clust) {
t_dist *row;
t_clustid *c;
int **nnb;
int i, j, k, cid, diff, max;
gmx_bool bChange;
real **mcpy = NULL;
if (rmsdcut < 0) {
rmsdcut = 10000;
}
/* First we sort the entries in the RMSD matrix row by row.
* This gives us the nearest neighbor list.
*/
</pre></article>
</slide>
<slide >
<hgroup>
<h2>Jarvis Patrick Clustering in Gromacs (Cont.)</h2>
<h3></h3>
</hgroup>
<article ><pre class="prettyprint" data-lang="c++">
// Five more pages of this
// You get the idea
// Also, how do we even use this function?
static void jarvis_patrick(int n1, real **mat, int M, int P,
real rmsdcut, t_clusters *clust);
</pre></article>
</slide>
<slide >
<hgroup>
<h2>Enter Data Science</h2>
<h3></h3>
</hgroup>
<article ><ul>
<li>Machine learning is mainstream now!</li>
<li>Thousands of experts are using machine learning approaches</li>
<li>Well-tested, performant, and facile implementations are available</li>
<li>Writing your own is not the way to go!<ul>
<li>E.g: Is clustering <em>that</em> special and MD-specific such that
we need our own custom algorithms and implementations?</li>
</ul>
</li>
</ul></article>
</slide>
<slide >
<hgroup>
<h2>MSMBuilder3: Design</h2>
<h3></h3>
</hgroup>
<article ><div style="float:right; margin-top:-100px">
<img src="figures/flow-chart.png" height="600">
</div>
<p>Builds on <a href="http://scikit-learn.org/stable/">scikit-learn</a> idioms:</p>
<ul>
<li>Everything is a <code>Model</code>.</li>
<li>Models are <code>fit()</code> on data.</li>
<li>Models learn <code>attributes_</code>.</li>
<li><code>Pipeline()</code> concatenate models.</li>
<li>Use best-practices (cross-validation)</li>
</ul>
<footer class="source">
<a href="http://msmbuilder.org/3.5.0/apipatterns.html">
http://msmbuilder.org/3.5.0/apipatterns.html
</a>
</footer></article>
</slide>
<slide >
<hgroup>
<h2>Everything is a <code>Model()</code>!</h2>
<h3></h3>
</hgroup>
<article ><pre class="prettyprint" data-lang="python">
>>> import msmbuilder.cluster
>>> clusterer = msmbuilder.cluster.KMeans(n_clusters=4)
>>> import msmbuilder.decomposition
>>> tica = msmbuilder.decomposition.tICA(n_components=3)
>>> import msmbuilder.msm
>>> msm = msmbuilder.msm.MarkovStateModel()
</pre>
<p>Hyperparameters go in the constructor.</p>
<footer class="source">
Actually, everything is a <code>sklearn.base.BaseEstimator()</code>
</footer></article>
</slide>
<slide >
<hgroup>
<h2>Models <code>fit()</code> data!</h2>
<h3></h3>
</hgroup>
<article ><pre class="prettyprint" data-lang="python">
>>> import msmbuilder.cluster
>>> trajectories = [np.random.normal(size=(100, 3))]
>>> clusterer = msmbuilder.cluster.KMeans(n_clusters=4, n_init=10)
>>> clusterer.fit(trajectories)
>>> clusterer.cluster_centers_
array([[-0.22340896, 1.0745301 , -0.40222902],
[-0.25410827, -0.11611431, 0.95394687],
[ 1.34302485, 0.14004818, 0.01130485],
[-0.59773874, -0.82508303, -0.95703567]])
</pre>
<p>Estimated parameters <em>always</em> have trailing underscores!</p></article>
</slide>
<slide >
<hgroup>
<h2><code>fit()</code> acts on lists of sequences</h2>
<h3></h3>
</hgroup>
<article ><pre class="prettyprint" data-lang="python">
>>> import msmbuilder.msm
>>> trajectories = [np.array([0, 0, 0, 1, 1, 1, 0, 0])]
>>> msm = msmbuilder.msm.MarkovStateModel()
>>> msm.fit(trajectories)
>>> msm.transmat_
array([[ 0.75 , 0.25 ],
[ 0.33333333, 0.66666667]])
</pre>
<p>This is different from sklearn, which uses 2D arrays.</p></article>
</slide>
<slide >
<hgroup>
<h2>Models <code>transform()</code> data!</h2>
<h3></h3>
</hgroup>
<article ><pre class="prettyprint" data-lang="python">
>>> import msmbuilder.cluster
>>> trajectories = [np.random.normal(size=(100, 3))]
>>> clusterer = msmbuilder.cluster.KMeans(n_clusters=8, n_init=10)
>>> clusterer.fit(trajectories)
>>> Y = clusterer.transform(trajectories)
[array([5, 6, 6, 0, 5, 5, 1, 6, 1, 7, 5, 7, 4, 2, 2, 2, 5, 3, 0, 0, 1, 3, 0,
5, 5, 0, 4, 0, 0, 3, 4, 7, 3, 5, 5, 5, 6, 1, 1, 0, 0, 7, 4, 4, 2, 6,
1, 4, 2, 0, 2, 4, 4, 5, 2, 6, 3, 2, 0, 6, 3, 0, 7, 7, 7, 0, 0, 0, 3,
3, 2, 7, 6, 7, 2, 5, 1, 0, 3, 6, 3, 2, 0, 5, 0, 3, 4, 2, 5, 4, 1, 5,
5, 4, 3, 3, 7, 2, 1, 4], dtype=int32)]
</pre>
<p>Moving the data-items from one "space" / representation into another.</p></article>
</slide>
<slide >
<hgroup>
<h2><code>Pipeline()</code> concatenates models!</h2>
<h3></h3>
</hgroup>
<article ><pre class="prettyprint" data-lang="python">
>>> import msmbuilder.cluster, msmbuilder.msm
>>> from sklearn.pipeline import Pipeline
>>> trajectories = [np.random.normal(size=(100, 3))]
>>> clusterer = msmbuilder.cluster.KMeans(n_clusters=2, n_init=10)
>>> msm = msmbuilder.msm.MarkovStateModel()
>>> pipeline = Pipeline([("clusterer", clusterer), ("msm", msm)])
>>> pipeline.fit(trajectories)
>>> msm.transmat_
array([[ 0.53703704, 0.46296296],
[ 0.53333333, 0.46666667]])
</pre>
<p>Data "flows" through transformations in the pipeline.</p></article>
</slide>
<slide >
<hgroup>
<h2>Loading Trajectories</h2>
<h3></h3>
</hgroup>
<article ><p>You can use MDTraj to load your trajectory files</p>
<pre class="prettyprint" data-lang="python">
>>> import glob
>>> import mdtraj as md
>>> filenames = glob.glob("./Trajectories/ala_*.h5")
>>> trajectories = [md.load(filename) for filename in filenames]
</pre>
<footer class="source">
Note: for big datasets, you can get fancy with <code><a href="http://mdtraj.org/latest/api/generated/mdtraj.iterload.html">md.iterload</a></code>.
</footer></article>
</slide>
<slide >
<hgroup>
<h2>Featurization</h2>
<h3></h3>
</hgroup>
<article ><p>Featurizers wrap MDTraj functions via the <code>transform()</code> function</p>
<div style="float:right;">
<img height=225 src=figures/rama.png />
</div>
<pre class="prettyprint" style="width:75%" data-lang="python">
>>> from msmbuilder.featurizer import DihedralFeaturizer
>>> from matplotlib.pyplot import hexbin, plot
>>> featurizer = DihedralFeaturizer(
... ["phi", "psi"], sincos=False)
>>> X = featurizer.transform(trajectories)
>>> phi, psi = np.rad2deg(np.concatenate(X).T)
>>> hexbin(phi, psi)
</pre>
<footer class="source">
<a href="http://msmbuilder.org/3.5.0/featurization.html">
http://msmbuilder.org/3.5.0/featurization.html
</a>
</footer></article>
</slide>
<slide >
<hgroup>
<h2>Featurization (Cont.)</h2>
<h3></h3>
</hgroup>
<article ><p>You can even combine featurizers with <code>FeatureSelector</code></p>
<pre class="prettyprint" data-lang="python">
>>> from msmbuilder.featurizer import DihedralFeaturizer, ContactFeaturizer
>>> from msmbuilder.feature_selection import FeatureSelector
>>> dihedrals = DihedralFeaturizer(
... ["phi", "psi"], sincos=True)
>>> contacts = ContactFeaturizer(scheme='ca')
>>> featurizer = FeatureSelector([('dihedrals', dihedrals),
... ('contacts', contacts)])
>>> X = featurizer.transform(trajectories)
</pre>
<footer class="source">
<a href="http://msmbuilder.org/3.5.0/featurization.html">
http://msmbuilder.org/3.5.0/featurization.html
</a>
</footer></article>
</slide>
<slide >
<hgroup>
<h2>Preprocessing</h2>
<h3></h3>
</hgroup>
<article ><p>Preprocessors normalize/whiten your data</p>
<pre class="prettyprint" data-lang="python">
>>> from msmbuilder.preprocessing import RobustScaler
>>> scaler = RobustScaler()
>>> Y = scaler.transform(X)
</pre>
<p>This is essential when combining different featurizers!</p>
<p>Also check out <code>MinMaxScaler</code> and <code>StandardScaler</code></p></article>
</slide>
<slide >
<hgroup>
<h2>Decomposition</h2>
<h3></h3>
</hgroup>
<article ><p>Reduce the dimensionality of your data</p>
<div style="float:right;">
<img width=275 src="http://msmbuilder.org/3.5.0/_images/tica_vs_pca.png"/>
<figcaption style="font-size: 50%; text-align: center;">
tICA finds the slowest degrees<br>of freedom in time-series data
</figcaption>
</div>
<pre class="prettyprint" style="width:75%" data-lang="python">
>>> from msmbuilder.decomposition import tICA
>>> tica = tICA(n_components=2, lagtime=5)
>>> Y = tica.fit_transform(X)
</pre>
<p>Also check out <code>PCA</code> and <code>SparseTICA</code></p>
<footer class="source">
<a href="http://msmbuilder.org/3.5.0/decomposition.html">
http://msmbuilder.org/3.5.0/decomposition.html
</a>
</footer></article>
</slide>
<slide >
<hgroup>
<h2>Markov State Models</h2>
<h3></h3>
</hgroup>
<article ><p>We offer two main flavors of MSM:</p>
<ul>
<li><code>MarkovStateModel</code> - Fits a first-order Markov model to a discrete-time integer labeled timeseries.</li>
<li><code>ContinuousTimeMSM</code> - Estimates a continuous rate matrix from discrete-time integer labeled timeseries.</li>
</ul>
<p>Each has a Bayesian version, which estimates the error associated with the model.</p>
<footer class="source">
MarkovStateModel: <a href="http://msmbuilder.org/3.5.0/msm.html">
http://msmbuilder.org/3.5.0/msm.html
</a>
<br>
ContinuousTimeMSM: <a href="http://msmbuilder.org/3.5.0/ratematrix.html">
http://msmbuilder.org/3.5.0/ratematrix.html
</a>
</footer></article>
</slide>
<slide >
<hgroup>
<h2>Hidden Markov Models</h2>
<h3></h3>
</hgroup>
<article ><div style="float: right;">
<img height=225" src="http://msmbuilder.org/3.5.0/_images/kde-vs-histogram.png"/>
<figcaption style="font-size: 50%; text-align: center;">KDE is to histogram as HMM is to MSM</figcaption>
</div>
<p>We also offer two types of HMMs:</p>
<ul>
<li><code>GaussianHMM</code> - Reversible Gaussian Hidden Markov Model L1-Fusion Regularization</li>
<li><code>VonMisesHMM</code> - Hidden Markov Model with von Mises Emissions</li>
</ul>
<p>HMMs are great for macrostate modeling!</p>
<footer class="source">
<a href="http://msmbuilder.org/3.5.0/hmm.html">
http://msmbuilder.org/3.5.0/hmm.html
</a>
</footer></article>
</slide>
<slide >
<hgroup>
<h2>Cross-Validation</h2>
<h3></h3>
</hgroup>
<article ><pre class="prettyprint" data-lang="python">
from sklearn.cross_validation import ShuffleSplit
cv = ShuffleSplit(len(trajectories), n_iter=5, test_size=0.5)
for fold, (train_index, test_index) in enumerate(cv):
train_data = [trajectories[i] for i in train_index]
test_data = [trajectories[i] for i in test_index]
model.fit(train_data)
model.score(test_data)
</pre>
<p>Also check out scikit-learn's <code>KFold</code>, <code>GridSearchCV</code> and <code>RandomizedSearchCV</code>.</p>
<footer class="source">
If you'd like to see how CV can be done with MSMs, see
<a href="http://msmbuilder.org/3.5.0/gmrq.html">
http://msmbuilder.org/3.5.0/gmrq.html
</a>
</footer></article>
</slide>
<slide >
<hgroup>
<h2>Command-line Tools</h2>
<h3></h3>
</hgroup>
<article ><p>We also offer an easy-to-use CLI for the API-averse</p>
<pre class="prettyprint" data-lang="shell">
$ msmb DihedralFeaturizer --top my_protein.pdb --trjs "*.xtc" \
--transformed diheds --out featurizer.pkl
$ msmb tICA -i diheds/ --out tica_model.pkl \
--transformed tica_trajs.h5 --n_components 4
$ msmb MiniBatchKMeans -i tica_trajs.h5 \
--transformed labeled_trajs.h5 --n_clusters 100
$ msmb MarkovStateModel -i labeled_trajs.h5 \
--out msm.pkl --lag_time 1
</pre>
<footer class="source">
<a href="http://msmbuilder.org/3.5.0/examples/Intro/Intro.cmd.html">
http://msmbuilder.org/3.5.0/examples/Intro/Intro.cmd.html
</a>
</footer></article>
</slide>
<slide >
<hgroup>
<h2>Related Projects</h2>
<h3></h3>
</hgroup>
<article ><p>We also maintain:</p>
<ul>
<li><a href="http://github.com/msmbuilder/osprey"><b>Osprey</b></a>- machine learning hyperparameter optimization</li>
<li><a href="http://github.com/msmbuilder/mdentropy"><b>MDEntropy</b></a> - entropy calculations for MD data</li>
</ul></article>
</slide>
<slide >
<hgroup>
<h2>Osprey</h2>
<h3></h3>
</hgroup>
<article ><p>Fully-automated, large-scale hyperparameter optimization</p>
<div style="text-align: center">
<img height=225 style="padding-bottom: 1em;" src="http://msmbuilder.org/osprey/development/_static/osprey.svg"/>
<figcaption>http://github.com/msmbuilder/osprey</figcaption>
</div>
<footer class="source">
Not just for MSMs!
</footer></article>
</slide>
<slide >
<hgroup>
<h2>Osprey: Estimator</h2>
<h3></h3>
</hgroup>
<article ><p>Define your model</p>
<pre class="prettyprint" data-lang="yaml">
estimator:
# The model/estimator to be fit.
eval_scope: msmbuilder
eval: |
Pipeline([
('featurizer', DihedralFeaturizer(types=['phi', 'psi'])),
('scaler', RobustScaler()),
('tica', tICA(n_components=2)),
('cluster', MiniBatchKMeans()),
('msm', MarkovStateModel(n_timescales=5, verbose=False)),
])
</pre></article>
</slide>
<slide >
<hgroup>
<h2>Osprey: Search Strategy</h2>
<h3></h3>
</hgroup>
<article ><p>Choose how to search over your hyperparameter space</p>
<pre class="prettyprint" data-lang="yaml">
strategy:
name: gp # or random, grid, hyperopt_tpe
params:
seeds: 50
</pre></article>
</slide>
<slide >
<hgroup>
<h2>Osprey: Search Space</h2>
<h3></h3>
</hgroup>
<article ><p>Select which hyperparameters to optimize</p>
<pre class="prettyprint" data-lang="yaml">
search_space:
featurizer__types:
choices:
- ['phi', 'psi']
- ['phi', 'psi', 'chi1']
type: enum
cluster__n_clusters:
min: 2
max: 1000
type: int
warp: log # search over log-space
</pre></article>
</slide>
<slide >
<hgroup>
<h2>Osprey: Cross-Validation</h2>
<h3></h3>
</hgroup>
<article ><p>Pick your favorite cross-validator</p>
<pre class="prettyprint" data-lang="yaml">
cv:
name: shufflesplit # Or kfold, loo, stratifiedshufflesplit, stratifiedkfold, fixed
params:
n_iter: 5
test_size: 0.5
</pre></article>
</slide>
<slide >
<hgroup>
<h2>Osprey: Dataset Loader</h2>
<h3></h3>
</hgroup>
<article ><p>Load your data, no matter what file type</p>
<pre class="prettyprint" data-lang="yaml">
dataset_loader:
# specification of the dataset on which to train the models.
name: mdtraj # Or msmbuilder, numpy, filename, joblib, sklearn_dataset, hdf5
params:
trajectories: ./fs_peptide/trajectory-*.xtc
topology: ./fs_peptide/fs-peptide.pdb
stride: 100
</pre></article>
</slide>
<slide >
<hgroup>
<h2>Osprey: Trials</h2>
<h3></h3>
</hgroup>
<article ><p>Save to a single SQL-like database, run on as many clusters as you'd like*</p>
<pre class="prettyprint" data-lang="yaml">
trials:
# path to a database in which the results of each hyperparameter fit
# are stored any SQL database is supported, but we recommend using
# SQLLite, which is simple and stores the results in a file on disk.
uri: sqlite:///osprey-trials.db
</pre>
<footer class="source">
*you'll still need to copy your data to each cluster, however
</footer></article>
</slide>
<slide >
<hgroup>
<h2>Osprey: Running a Job</h2>
<h3></h3>
</hgroup>
<article ><p>Simple command-line interface, easy to run on any cluster</p>
<pre class="prettyprint" data-lang="SHELL">
$ osprey worker -n 100 config.yaml
...
----------------------------------------------------------------------
Beginning iteration 10 / 100
----------------------------------------------------------------------
Loading trials database: sqlite:///trials.db...
History contains: 9 trials
Choosing next hyperparameters with gp...
{'tica__n_components': 2, 'tica__lag_time': 180, 'cluster__n_clusters': 36}
(gp took 0.000 s)
...
Success! Model score = 4.214510
(best score so far = 4.593165)
</pre></article>
</slide>
<slide >
<hgroup>
<h2>Osprey: Real-Time Analytics</h2>
<h3></h3>
</hgroup>
<article ><p>Osprey also makes it easy to create interactive dashboards</p>
<pre class="prettyprint" data-lang="SHELL">
$ osprey plot config.yaml
</pre>
<div style="text-align: center">
<img height=350 style="padding-top: .5em;" src="figures/osprey-dash.png"/>
<img height=350 style="padding-top: .5em;" src="figures/osprey-dash2.png"/>
</div></article>
</slide>
<slide class="thank-you-slide segue nobackground">
<!-- <aside class="gdbar right"><img src="images/google_developers_icon_128.png"></aside> -->
<article class="flexbox vleft auto-fadein">
<h2></h2>
<p></p>
</article>
<p data-config-contact class="auto-fadein"></p>
</p>
</slide>
<slide class="backdrop"></slide>
</slides>
<script>
var _gaq = _gaq || [];
_gaq.push(['_setAccount', 'UA-XXXXXXXX-1']);
_gaq.push(['_trackPageview']);
(function() {
var ga = document.createElement('script'); ga.type = 'text/javascript'; ga.async = true;
ga.src = ('https:' == document.location.protocol ? 'https://ssl' : 'http://www') + '.google-analytics.com/ga.js';
var s = document.getElementsByTagName('script')[0]; s.parentNode.insertBefore(ga, s);
})();
</script>
<!--[if IE]>
<script src="http://ajax.googleapis.com/ajax/libs/chrome-frame/1/CFInstall.min.js"></script>
<script>CFInstall.check({mode: 'overlay'});</script>
<![endif]-->
</body>
</html>