README.org 40.8 KB
Newer Older
Maude Le Jeune's avatar
Maude Le Jeune committed
1 2 3
#+TITLE: The Pipelet Readme
#+STYLE: <link rel="stylesheet" type="text/css" href="org.css" />

Maude Le Jeune's avatar
Maude Le Jeune committed
4 5
Pipelet is a free framework allowing for the creation, execution and
browsing of scientific data processing pipelines. It provides:
6

Marc Betoule's avatar
Marc Betoule committed
7 8 9 10 11
+ easy chaining of interdependent elementary tasks,
+ web access to data products,
+ branch handling,
+ automated distribution of computational tasks.

12 13 14
Both engine and web interface are written in Python. As Pipelet is all
about chaining processing written in Python or using Python as a glue
language, prior knowledge of this language is required.
Marc Betoule's avatar
Marc Betoule committed
15 16 17 18 19

* Tutorial
** Introduction
*** Why using pipelines

Maude Le Jeune's avatar
Maude Le Jeune committed
20
The pipeline mechanism allows you to apply a sequence of processing
Marc Betoule's avatar
Marc Betoule committed
21 22 23 24 25 26 27 28
steps to your data, in a way that the input of each process is the
output of the previous one. Making visible these different processing
steps, in the right order, is essential in data analysis to keep track
of what you did, and make sure that the whole processing remains
consistent.

*** How it works

Maude Le Jeune's avatar
Maude Le Jeune committed
29 30 31
Pipelet is based on the possibility to save on disk every intermediate
input or output of your pipeline, which is usually not a strong
constraint but offers a lot of benefits. It means that you can stop
32
the processing whenever you want, and start it again without
Marc Betoule's avatar
Marc Betoule committed
33 34 35 36 37
recomputing the whole thing: you just take the last products you have
on disk, and continue the processing where it stopped. This logic is
interesting when the computation cost is higher than the cost of disk
space required by intermediate products.

Maude Le Jeune's avatar
readme  
Maude Le Jeune committed
38 39 40 41 42 43 44 45
In addition, the Pipelet engine has been designed to
process *data* *sets*. It takes advantage of the parallelisation
opportunity that comes with data which share the same structure (data
arrays), to dispatch the computational tasks on parallel architecture.
The data dependency scheme is also used to save CPU time, and allows
to handle very big data sets processing.


Marc Betoule's avatar
Marc Betoule committed
46 47
*** The Pipelet functionalities

Maude Le Jeune's avatar
Maude Le Jeune committed
48 49
Pipelet is a free framework which helps you : 
+ to write and manipulate pipelines with any dependency scheme, 
Maude Le Jeune's avatar
readme  
Maude Le Jeune committed
50 51 52
+ to keep track of what processing has been applied to your data and perform comparisons,
+ to carry pipelines source code from development to production and
  adapt to different hardware and software architectures.
Marc Betoule's avatar
Marc Betoule committed
53

Maude Le Jeune's avatar
Maude Le Jeune committed
54

55 56 57 58
*** What's new in v1.1
- The glob_seg behavior has been modified for coherence, convenience,
  and performance sake. See [[*The%20segment%20environment][The segment environment]].

Marc Betoule's avatar
Marc Betoule committed
59
** Getting started
Maude Le Jeune's avatar
Maude Le Jeune committed
60 61 62 63 64 65
*** Pipelet installation 
**** Dependencies 

+ Running the Pipelet engine requires Python >= 2.6.

+ The web interface of Pipelet requires the installation of the
66
  cherrypy3 Python module (on Debian: aptitude install python-cherrypy3).
Maude Le Jeune's avatar
Maude Le Jeune committed
67

68
You may find useful to install some generic scientific tools that nicely interact with Pipelet: 
Maude Le Jeune's avatar
Maude Le Jeune committed
69 70 71
+ numpy
+ matplotlib
+ latex 
Marc Betoule's avatar
Marc Betoule committed
72

Maude Le Jeune's avatar
Maude Le Jeune committed
73
**** Getting Pipelet
Marc Betoule's avatar
Marc Betoule committed
74

Maude Le Jeune's avatar
readme  
Maude Le Jeune committed
75 76 77 78 79 80 81 82 83 84 85
***** Software status

The first version of the software is currently in the process of
stabilisation.  The Pipelet engine has now reached the level of
desired sophistication.  On the other hand, the user interface has
been developped in a minimalist way. It includes the main
functionalities but with a design which could and will be more user
friendly. 


***** Getting last pipelet version
Marc Betoule's avatar
Marc Betoule committed
86

Maude Le Jeune's avatar
Maude Le Jeune committed
87
=git clone -o v1.0 git://gitorious.org/pipelet/pipelet.git=
Marc Betoule's avatar
Marc Betoule committed
88

Maude Le Jeune's avatar
Maude Le Jeune committed
89
**** Installing Pipelet
Marc Betoule's avatar
Marc Betoule committed
90

Marc Betoule's avatar
Marc Betoule committed
91
=sudo python setup.py install=
Marc Betoule's avatar
Marc Betoule committed
92

93
*** Running a simple test pipeline
Marc Betoule's avatar
Marc Betoule committed
94 95 96

1. Run the test pipeline

Maude Le Jeune's avatar
Maude Le Jeune committed
97
   =cd test/first_test=
Marc Betoule's avatar
Marc Betoule committed
98

Maude Le Jeune's avatar
Maude Le Jeune committed
99
   =python main.py=
Marc Betoule's avatar
Marc Betoule committed
100 101 102

2. Add this pipeline to the web interface

Maude Le Jeune's avatar
Maude Le Jeune committed
103
   =pipeweb track test ./.sqlstatus=
Marc Betoule's avatar
Marc Betoule committed
104

105
3. Set up an account in the access control list and launch the web server
Marc Betoule's avatar
Marc Betoule committed
106

Maude Le Jeune's avatar
Maude Le Jeune committed
107
   =pipeutils -a username -l 2 .sqlstatus=
Marc Betoule's avatar
Marc Betoule committed
108

Maude Le Jeune's avatar
Maude Le Jeune committed
109
   =pipeweb start=
Marc Betoule's avatar
Marc Betoule committed
110 111 112 113

4. You should be able to browse the result on the web page
   http://localhost:8080

114 115
*** Getting a new pipe framework

Maude Le Jeune's avatar
Maude Le Jeune committed
116
To get a new pipeline framework, with example main and segment scripts : 
117

Marc Betoule's avatar
Marc Betoule committed
118
=pipeutils -c pipename=
119

Maude Le Jeune's avatar
Maude Le Jeune committed
120 121 122
This command ends up with the creation of directory named pipename wich contains: 
+ a main script (named main.py) providing functionnalities to execute
  your pipeline in various modes (debug, parallel, batch mode, ...)
123
+ an example of segment script (=default.py=) which illustrates
Maude Le Jeune's avatar
Maude Le Jeune committed
124
  the pipelet utilities with comments. 
125

Maude Le Jeune's avatar
Maude Le Jeune committed
126
The next section describes those two files in more details. 
127

Maude Le Jeune's avatar
Maude Le Jeune committed
128
** Writing Pipes
Marc Betoule's avatar
Marc Betoule committed
129 130 131 132 133 134 135 136 137 138
*** Pipeline architecture

The definition of a data processing pipeline consists in :
+ a succession of python scripts, called segments, coding each step
  of the actual processing.
+ a main script that defines the dependency scheme between segments,
  and launch the processing.

The dependencies between segments must form a directed acyclic
graph. This graph is described by a char string using a subset of the
139
graphviz dot language (http://www.graphviz.org). For example the string:
Marc Betoule's avatar
Marc Betoule committed
140

Marc Betoule's avatar
Marc Betoule committed
141
#+begin_src python
Marc Betoule's avatar
Marc Betoule committed
142
"""
Marc Betoule's avatar
Marc Betoule committed
143 144 145
a -> b -> d;
c -> d;
c -> e;
Marc Betoule's avatar
Marc Betoule committed
146
"""
Marc Betoule's avatar
Marc Betoule committed
147 148 149 150
#+end_src

defines a pipeline with 5 segments ={"a", "b", "c", "d", "e"}=. The
relation =a->b= ensures that the processing of the segment "a" will be
151
done before the processing of its 'child' segment =b=. Also the output
Marc Betoule's avatar
Marc Betoule committed
152 153 154
of =a= will be fed as input for =b=. In the given example, the node
=d= has two parents =b= and =c=. Both will be executed before =d=. As
their is no relation between =b= and =c= which of the two will be
Marc Betoule's avatar
Marc Betoule committed
155 156
executed first is not defined.

Marc Betoule's avatar
Marc Betoule committed
157 158 159
When executing the segment =seg=, the engine looks for a python script
named =seg.py=. If not found, it looks iteratively for script files
named =se.py= and =s.py=. This way, different segments of the pipeline
Maude Le Jeune's avatar
Maude Le Jeune committed
160
can share the same code, if they are given a name with a common root
161
(this mechanism is useful to write generic segment and is completed by
Maude Le Jeune's avatar
Maude Le Jeune committed
162
the hooking system, described in the advanced usage section). The code
Marc Betoule's avatar
Marc Betoule committed
163 164
is then executed in a specific namespace (see below [[*The%20segment%20environment][The execution
environment]]).
Maude Le Jeune's avatar
Maude Le Jeune committed
165

Marc Betoule's avatar
Marc Betoule committed
166 167
*** The Pipeline object

168
Practically, the creation of a Pipeline object requires 3 arguments:
Marc Betoule's avatar
Marc Betoule committed
169

Marc Betoule's avatar
Marc Betoule committed
170
#+begin_src python
171
from pipelet.pipeline import Pipeline
Marc Betoule's avatar
Marc Betoule committed
172 173
P = Pipeline(pipedot, codedir="./", prefix="./")
#+end_src
Marc Betoule's avatar
Marc Betoule committed
174

Marc Betoule's avatar
Marc Betoule committed
175 176 177 178

- =pipedot= is the string description of the pipeline
- =codedir= is the path where the segment scripts can be found
- =prefix=  is the path to the data repository (see below [[*Hierarchical%20data%20storage][Hierarchical data storage]])
Marc Betoule's avatar
Marc Betoule committed
179

180
It is possible to output the graphviz representation of the pipeline
181 182
(requires the installation of graphviz). First, save the graph string
into a .dot file with the pipelet function:
Marc Betoule's avatar
Marc Betoule committed
183 184

#+begin_src python
Maude Le Jeune's avatar
Maude Le Jeune committed
185
P.to_dot('pipeline.dot')
Marc Betoule's avatar
Marc Betoule committed
186
#+end_src
Maude Le Jeune's avatar
Maude Le Jeune committed
187 188 189

Then, convert it to an image file with the dot command: 

Marc Betoule's avatar
Marc Betoule committed
190
=dot -Tpng -o pipeline.png pipeline.dot=
Maude Le Jeune's avatar
Maude Le Jeune committed
191

192
*** Dependencies between segments and data parallelism
Marc Betoule's avatar
Marc Betoule committed
193

Marc Betoule's avatar
Marc Betoule committed
194 195 196 197 198
The modification of the code of one segment will trigger its
recalculation and the recalculation of all the segments which
depend on it.

The output of a segment is a list of python objects. If a segment as
Maude Le Jeune's avatar
Maude Le Jeune committed
199 200 201 202 203 204 205 206 207 208
no particular output this list can be empty and do not need to be
specified. Elements of the list are allowed to be any kind of
pickleable python objects. However, a good practice is to fill the
list with the minimal set of characteristics relevant to describe the
output of the segment and to defer the storage of the data to
appropriate structures and file formats. For example, a segment which
performs computation on large images could virtually pass the results
of its computation to the following segment using the output list. It
is a better practice to store the resulting image in a dedicated file
and to pass in the list only the information allowing a non ambiguous
209
identification of this file (like its name or part of it).
Maude Le Jeune's avatar
Maude Le Jeune committed
210 211 212 213

The input of a child segment is taken in a set build from the output
lists of its parents. The content of the input set is actually tunable
using the multiplex directive (see below). However the simplest and
214
default behavior of the engine is to form the Cartesian product of
Maude Le Jeune's avatar
Maude Le Jeune committed
215 216
the output list of its parent.

217
To illustrate this behavior let us consider the following pipeline,
Maude Le Jeune's avatar
Maude Le Jeune committed
218 219
build from three segments:

Marc Betoule's avatar
Marc Betoule committed
220 221 222 223 224 225
#+begin_src python
"""
knights -> melt;
quality -> melt;
"""
#+end_src
Maude Le Jeune's avatar
Maude Le Jeune committed
226 227 228 229

and assume that the respective output lists of segments knights and
quality are:

Marc Betoule's avatar
Marc Betoule committed
230 231 232
#+begin_src python
["Lancelot", "Galahad"]
#+end_src
Maude Le Jeune's avatar
Maude Le Jeune committed
233
and:
Marc Betoule's avatar
Marc Betoule committed
234 235 236
#+begin_src python
['the Brave', 'the Pure']
#+end_src
Maude Le Jeune's avatar
Maude Le Jeune committed
237

238
The Cartesian product of the previous set is:
Marc Betoule's avatar
Marc Betoule committed
239
#+begin_src python
Maude Le Jeune's avatar
Maude Le Jeune committed
240 241
 [('Lancelot','the Brave'), ('Lancelot,'the Pure'), ('Galahad','the Brave'), ('Galahad','the
Pure')]
Marc Betoule's avatar
Marc Betoule committed
242
#+end_src
Maude Le Jeune's avatar
Maude Le Jeune committed
243

Marc Betoule's avatar
Marc Betoule committed
244
Four instances of segment =melt= will thus be run, each one receiving
Maude Le Jeune's avatar
Maude Le Jeune committed
245 246 247
as input one of the four 2-tuples.

At the end of the execution of all the instances of a segment, their
Marc Betoule's avatar
Marc Betoule committed
248
output lists are concatenated. If the action of segment =melt= is to
Maude Le Jeune's avatar
Maude Le Jeune committed
249
concatenate the two strings he receives separated by a space, the
Marc Betoule's avatar
Marc Betoule committed
250
final output set of segment =melt= will be: 
Maude Le Jeune's avatar
Maude Le Jeune committed
251

Marc Betoule's avatar
Marc Betoule committed
252 253 254
#+begin_src python
[('Lancelot the Brave'), ('Lancelot the Pure'), ('Galahad the Brave'), ('Galahad the Pure')].
#+end_src
Marc Betoule's avatar
Marc Betoule committed
255

Marc Betoule's avatar
Marc Betoule committed
256
This default behavior can be altered by specifying a =#multiplex=
Marc Betoule's avatar
Marc Betoule committed
257 258
directive in the commentary of the segment code. See section [[*Multiplex%20directive][Multiplex
directive]] for more details.
259

260 261 262 263
As the segment execution order is not uniquely determined by the pipe
scheme (several path may exists), it is not possible to retrieve an
ordered input tuple. To overcome this issue, segment inputs are
dictionaries, with keywords matching parent segment names.  In the
Marc Betoule's avatar
Marc Betoule committed
264
above example, one can read =melt= inputs using:
265

Marc Betoule's avatar
Marc Betoule committed
266
#+begin_src python
267 268
k = seg_input["knights"]
q = seg_input["quality"]
Marc Betoule's avatar
Marc Betoule committed
269
#+end_src
270

Marc Betoule's avatar
Marc Betoule committed
271
See section [[*The%20segment%20environment]['The segment environment']] for more details.
272

Marc Betoule's avatar
Marc Betoule committed
273
*** Orphan segments
Marc Betoule's avatar
Marc Betoule committed
274

275 276 277 278 279
By default, orphan segments (segments without parents) have no input
argument (an empty list), and therefore are executed once without
input. The possibility is offered to feed input to an orphan segment
by pushing a list into the output set of an hypothetic ('phantom')
parent. If P is an instance of the pipeline object, this is done by:
280

Marc Betoule's avatar
Marc Betoule committed
281
#+begin_src python
282
P.push (segname=[1,2,3])
Marc Betoule's avatar
Marc Betoule committed
283
#+end_src
284 285

From the segment environment, inputs can be retrieve from the
Marc Betoule's avatar
Marc Betoule committed
286
usual dictionary, using the keyword =segnamephantom=. 
287

Marc Betoule's avatar
Marc Betoule committed
288
#+begin_src python
289
id = seg_input['segnamephantom']
Marc Betoule's avatar
Marc Betoule committed
290
#+end_src
291
or
Marc Betoule's avatar
Marc Betoule committed
292
#+begin_src python
293
id = seg_input.values()[0]
Marc Betoule's avatar
Marc Betoule committed
294
#+end_src
295

296 297 298 299 300 301 302
In this scheme, it is important to uniquely identify the child tasks
of the orphan segment by setting a dedicated output.

#+begin_src python
seg_output = id
#+end_src

Marc Betoule's avatar
Marc Betoule committed
303
See section [[*The%20segment%20environment][The segment environment]] for more details.
304

Marc Betoule's avatar
Marc Betoule committed
305 306
*** Hierarchical data storage

307 308 309 310 311 312
The framework provides versioning of your data and easy access through
the web interface. It also keep track of the code, of the execution
logs, and various meta-data of the processing. Of course, you remain
able to bypass the hierarchical storage and store your actual data
elsewhere, but you will loose the benefit of automated versioning
which proves to be quite convenient.
Marc Betoule's avatar
Marc Betoule committed
313 314

The storage is organized as follows:
315 316 317

- all pipeline instances are stored below a root which corresponds to
  the prefix parameter of the Pipeline object. 
Marc Betoule's avatar
Marc Betoule committed
318
      =/prefix/=
319
- all segment meta data are stored below a root which name corresponds
320
  to a unique hash computed on the segment code and its dependencies.
Marc Betoule's avatar
Marc Betoule committed
321
      =/prefix/segname_YFLJ65/=
322 323 324 325 326 327 328
- Segment's meta data are: 
  - a copy of the segment python script
  - a copy of all segment hook scripts
  - a parameter file (.args) which contains segment parameters value
  - a meta data file (.meta) which contains some extra meta data
- all segment instances data and meta data are stored in a specific subdirectory
  which name corresponds to a string representation of its input
329 330
  prefix by its identifier number
  	=/prefix/segname_YFLJ65/data/1_a/=
331
- if there is a single segment instance, then data are stored in
Marc Betoule's avatar
Marc Betoule committed
332
       =/prefix/segname_YFLJ65/data/=
333 334
- If a segment has at least one parent, its root will be located below
  one of its parent's one : 
Marc Betoule's avatar
Marc Betoule committed
335
       =/prefix/segname_YFLJ65/segname2_PLMBH9/=
336
- etc...
Maude Le Jeune's avatar
Maude Le Jeune committed
337
  
338 339 340 341 342 343
While the hierarchical storage makes easy the storing and handling of
many different data with different versions, it can make the manual
navigation in the data less convenient. Here comes the role of the [[*Browsing%20Pipes][web
interface]] (among other advantages, like distant access to the data,
tagging...).

Marc Betoule's avatar
Marc Betoule committed
344 345 346 347 348
*** The segment environment

The segment code is executed in a specific environment that provides:

1. access to the segment input and output
349 350 351
   - =get_input(seg)=: return the input coming from segment seg. If no
     segment specified, take the first. This utility replaces the
     seg_input variable which type could vary as described below.
352

353
   - =seg_input=:  this variable is a dictionary containing the input of the segment.
Marc Betoule's avatar
Marc Betoule committed
354
     In the general case, =seg_input= is a python dictionary which
355
     contains as many keywords as parent segments. In the case of orphan
Marc Betoule's avatar
Marc Betoule committed
356 357
     segment, the keyword used is suffixed by the =phantom= word. 
     One exception to this is coming from the use of the =group_by=
358
     directive, which alters the origin of the inputs. In this case,
359
     =seg_input= contains the resulting class elements. 
360 361 362
     
   - =set_output(o)=: set the segment output as a list. If o is not a
     list, set a list of one element o. 
363

Marc Betoule's avatar
Marc Betoule committed
364
   - =seg_output=: this variable has to be a list. 
Marc Betoule's avatar
Marc Betoule committed
365

366
2. Functionalities to use the automated hierarchical data storage system.
367 368 369 370 371 372 373 374 375 376
   - =get_data_fn(basename)=: complete the filename with the path to
     the working directory. 
   - =glob_parent(regexp, segs)=: Return the list of filename matching
        the pattern y in the data directory of direct parent tasks. It
        is possible to search only in a specific segment list segs.
   - =glob_seg(seg, regexp)=: Return the list of filename matching the
        pattern y in the data directory of parent segment x (all task
        directories are searched, independantly of whether the file
        comes from a task related to the current task). Its usage
        should be limited as it:
377 378 379
        - potentially breaks the dependancy scheme.
        - may hurt performances as all task directories of the segment
          x will be searched.
Marc Betoule's avatar
Marc Betoule committed
380
   - =get_tmp_fn()=: return a temporary filename.
Marc Betoule's avatar
Marc Betoule committed
381

382
3. Functionalities to use the automated parameters handling
383 384 385
   - =save(lst)=: save the listed parameters of the segment in the meta data.
   - =expose(lst)=: expose the listed parameters from the web interface
   - =load(seg, globals(), lst)=: retrieve parameters from the meta data.
386 387

4. Various convenient functionalities
388
   - =save_products(filename, globals(), lst_par)=: use pickle to save a
389
     part of the current namespace.
390
   - =load_products(filename, globals(), lst_par)=: update the namespace by
Marc Betoule's avatar
Marc Betoule committed
391
     unpickling requested object from the file.
Marc Betoule's avatar
Marc Betoule committed
392 393 394
   - =logged_subprocess(lst_args)=: execute a subprocess and log its
     output in =processname.log= and =processname.err=.
   - =logger= is a standard =logging.Logger= object that can be used to
Betoule Marc's avatar
Betoule Marc committed
395
     log the processing
Marc Betoule's avatar
Marc Betoule committed
396

397
5. Hooking support 
Marc Betoule's avatar
Marc Betoule committed
398 399
   Pipelet enables you to write reusable generic
   segments by providing a hooking system via the hook function.
Marc Betoule's avatar
Marc Betoule committed
400 401 402
   =hook(hookname, globals())=: execute Python script =segname_hookname.py= and update the namespace.
   See the section [[*The%20hooking%20system][Hooking system]] for more details.

Marc Betoule's avatar
Marc Betoule committed
403

404
*** Writing a first pipeline
Marc Betoule's avatar
Marc Betoule committed
405

406 407 408
We are now in the position to write a complete simple pipeline. Let us
consider the knights example and write the beginning of the main file
=main.py= describing the pipeline:
Maude Le Jeune's avatar
Maude Le Jeune committed
409

Marc Betoule's avatar
Marc Betoule committed
410
#+begin_src python
411 412 413 414 415 416 417 418
  from pipelet.pipeline import Pipeline
  
  pipedot = """
  knights -> melt;
  quality -> melt;
  """
  
  P = Pipeline(pipedot, code_dir='./',prefix='./')  
Marc Betoule's avatar
Marc Betoule committed
419
#+end_src
Maude Le Jeune's avatar
Maude Le Jeune committed
420

421 422 423
Now, we create the 3 segment files =knights.py=, =quality.py= and
=melt.py=. The only action we expect from segment knights is simply to
provide a list of knights. Its code is very simple:
Marc Betoule's avatar
Marc Betoule committed
424
#+begin_src python
425
  seg_output =  ["Lancelot", "Galahad"]
Marc Betoule's avatar
Marc Betoule committed
426
#+end_src
Maude Le Jeune's avatar
Maude Le Jeune committed
427

428
Same thing for the segment quality:
Marc Betoule's avatar
Marc Betoule committed
429
#+begin_src python
430
  seg_output = ['the Brave', 'the Pure']  
Marc Betoule's avatar
Marc Betoule committed
431
#+end_src
Maude Le Jeune's avatar
Maude Le Jeune committed
432

433 434
As explained, the segment melt will be executed four times. We expect
from it to concatenate its input and write the result into a file, so the code is:
Marc Betoule's avatar
Marc Betoule committed
435
#+begin_src python
436 437 438 439
  knight, quality = seg_input['knights'], seg_input['quality']
  f = open(get_data_fn('result.txt'), 'w')
  f.write(knight + ' ' + quality+'\n')
  f.close()  
Marc Betoule's avatar
Marc Betoule committed
440
#+end_src
Maude Le Jeune's avatar
Maude Le Jeune committed
441

442 443
We need to complete the main file so that it takes care of the
execution ([[*Running%20Pipes][see Running Pipes for more explainations]]):
Marc Betoule's avatar
Marc Betoule committed
444
#+begin_src python
445 446 447 448 449 450 451 452 453 454
  from pipelet.pipeline import Pipeline
  from pipelet.launchers import launch_interactive
  pipedot = """
  knights -> melt;
  quality -> melt;
  """
  
  P = Pipeline(pipedot, code_dir='./',prefix='./')
  w,t = launch_interactive(P)
  w.run()
Marc Betoule's avatar
Marc Betoule committed
455
#+end_src
Maude Le Jeune's avatar
Maude Le Jeune committed
456

457 458 459 460 461 462 463 464
The execution of the main file will run this simple example in the
'interactive' mode provided for debugging purposes. You may add a
knight in the list to see only the required part recomputed. More
complete examples are described in the [[*The%20example%20pipelines][example pipelines]] section. The
two remaining sections of the tutorial explain how to use execution
mode that enable to exploitation of data parallelism (in this case
running the four independent instances of the melt segment in
parallel), and how to provide web access to the results.
Maude Le Jeune's avatar
Maude Le Jeune committed
465

466 467 468 469
*** The exemple pipelines
**** fft
**** cmb

Marc Betoule's avatar
Marc Betoule committed
470
** Running Pipes
471
   
472 473
*** The sample main file

474
A sample main file is made available when creating a new Pipelet
475 476
framework. It is copied from the reference file: 

Marc Betoule's avatar
Marc Betoule committed
477
=pipelet/pipelet/static/main.py=
478 479 480 481 482 483 484

This script illustrates various ways of running pipes. It describes
the different parameters, and also, how to write a
main python script can be used as any binary from the command line
(including options parsing). 

*** Common options
485
    Some options are common to each running modes.
486 487 488 489
**** log level

The logging system is handle by the python logging facility module. 
This module defines the following log levels : 
Marc Betoule's avatar
Marc Betoule committed
490 491 492 493 494
+ =DEBUG=
+ =INFO=
+ =WARNING=
+ =ERROR=
+ =CRITICAL=
495

496
All logging messages are saved in the different Pipelet log files,
497 498 499 500 501
available from the web interface (rotating file logging).  It is also
possible to print those messages on the standard output (stream
logging), by setting the desired log level in the launchers options:
For example: 

Marc Betoule's avatar
Marc Betoule committed
502
#+begin_src python
503 504
import logging
launch_process(P, N,log_level=logging.DEBUG)
Marc Betoule's avatar
Marc Betoule committed
505
#+end_src
506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525

If set to 0, stream logging will be disable. 

**** matplotlib

The matplotlib documentation says: 

"Many users report initial problems trying to use maptlotlib in web
application servers, because by default matplotlib ships configured to
work with a graphical user interface which may require an X11
connection. Since many barebones application servers do not have X11
enabled, you may get errors if you don’t configure matplotlib for use
in these environments. Most importantly, you need to decide what kinds
of images you want to generate (PNG, PDF, SVG) and configure the
appropriate default backend. For 99% of users, this will be the Agg
backend, which uses the C++ antigrain rendering engine to make nice
PNGs. The Agg backend is also configured to recognize requests to
generate other output formats (PDF, PS, EPS, SVG). The easiest way to
configure matplotlib to use Agg is to call:

Marc Betoule's avatar
Marc Betoule committed
526
=matplotlib.use('Agg')=
527

528 529 530 531
The =matplotlib= and =matplotlib_interactive= options turn the
matplotlib backend to Agg in order to allow the execution in
non-interactive environment. The two options affects independently the
non interactive execution mode and the interactive mode.
532

Marc Betoule's avatar
Marc Betoule committed
533
Those two options are set to =True= by default in the sample main
534 535 536
script. Setting them to False deactivate this behavior for pipelines
that make no use of matplotlib (and prevents the raise of an exception
if matplotlib is not even available).
537

Marc Betoule's avatar
Marc Betoule committed
538
*** The interactive mode
Marc Betoule's avatar
Marc Betoule committed
539
This mode has been designed to ease debugging. If =P= is an instance of
540
the pipeline object, the syntax reads :
Marc Betoule's avatar
Marc Betoule committed
541

Marc Betoule's avatar
Marc Betoule committed
542
#+begin_src python
Marc Betoule's avatar
Marc Betoule committed
543 544 545
from pipelet.launchers import launch_interactive
w, t = launch_interactive(P)
w.run()
Marc Betoule's avatar
Marc Betoule committed
546
#+end_src
Marc Betoule's avatar
Marc Betoule committed
547 548 549 550

In this mode, each tasks will be computed in a sequential way. 
Do not hesitate to invoque the Python debugger from IPython : %pdb

Maude Le Jeune's avatar
Maude Le Jeune committed
551
To use the interactive mode, run: 
Marc Betoule's avatar
Marc Betoule committed
552
=main.py -d=
553

Marc Betoule's avatar
Marc Betoule committed
554
*** The process mode
555 556 557 558
In this mode, one can run simultaneous tasks (if the pipe scheme
allows to do so). 
The number of subprocess is set by the N parameter : 

Marc Betoule's avatar
Marc Betoule committed
559
#+begin_src python
560 561
from pipelet.launchers import launch_process
launch_process(P, N)
Marc Betoule's avatar
Marc Betoule committed
562
#+end_src
563

Maude Le Jeune's avatar
Maude Le Jeune committed
564
To use the process mode, run: 
Marc Betoule's avatar
Marc Betoule committed
565
=main.py=
Maude Le Jeune's avatar
Maude Le Jeune committed
566
or
Marc Betoule's avatar
Marc Betoule committed
567
=main.py -p 4=
Maude Le Jeune's avatar
Maude Le Jeune committed
568

Marc Betoule's avatar
Marc Betoule committed
569
*** The batch mode
570 571 572
In this mode, one can submit some batch jobs to execute the tasks. 
The number of job is set by the N parameter : 

Marc Betoule's avatar
Marc Betoule committed
573
#+begin_src python
574 575
from pipelet.launchers import launch_pbs
launch_pbs(P, N , address=(os.environ['HOST'],50000))
Marc Betoule's avatar
Marc Betoule committed
576
#+end_src
Marc Betoule's avatar
Marc Betoule committed
577

578 579 580 581 582 583 584 585 586 587 588
It is possible to specify some job submission options like: 
+ job name 
+ job header: this string is prepend to the PBS job scripts. You may
  want to add some environment specific paths. Log and error files are
  automatically handled by the pipelet engine, and made available from
  the web interface. 
+ cpu time: syntax is: "hh:mm:ss"

The 'server' option can be disable to add some workers to an existing
scheduler.

Maude Le Jeune's avatar
Maude Le Jeune committed
589
To use the batch mode, run: 
Marc Betoule's avatar
Marc Betoule committed
590
=main.py -b=
Maude Le Jeune's avatar
Maude Le Jeune committed
591 592 593

to start the server, and: 

Marc Betoule's avatar
Marc Betoule committed
594
=main.py -a 4=
595

Maude Le Jeune's avatar
Maude Le Jeune committed
596
to add 4 workers. 
Maude Le Jeune's avatar
Maude Le Jeune committed
597

598

Marc Betoule's avatar
Marc Betoule committed
599
** Browsing Pipes
600 601 602 603 604
*** The pipelet webserver and ACL

The pipelet webserver allows the browsing of multiple pipelines. 
Each pipeline has to be register using : 

Marc Betoule's avatar
Marc Betoule committed
605
=pipeweb track <shortname> sqlfile=
606

607 608
To remove a pipeline from the tracked list, use : 

Marc Betoule's avatar
Marc Betoule committed
609
=pipeweb untrack <shortname>=
610

611 612 613 614
As the pipeline browsing implies a disk parsing, some basic security
has to be set also. All users have to be register with a specific access
level (1 for read-only access, and 2 for write access).  

Marc Betoule's avatar
Marc Betoule committed
615
=pipeutils -a <username> -l 2 sqlfile=
Marc Betoule's avatar
Marc Betoule committed
616

617 618
To remove a user from the user list: 

Marc Betoule's avatar
Marc Betoule committed
619
=pipeutils -d <username> sqlfile=
620

621 622
Start the web server using : 

Marc Betoule's avatar
Marc Betoule committed
623
=pipeweb start=
624 625

Then the web application will be available on the web page http://localhost:8080
Marc Betoule's avatar
Marc Betoule committed
626

627 628
To stop the web server : 

Marc Betoule's avatar
Marc Betoule committed
629
=pipeweb stop=
630

Marc Betoule's avatar
Marc Betoule committed
631
*** The web application
632

633
In order to ease the comparison of different processing, the web
634 635 636 637
interface displays various views of the pipeline data : 

**** The index page 

638
The index page displays a tree view of all pipeline instances. Each
639 640 641 642 643 644
segment may be expand or reduce via the +/- buttons.  

The parameters used in each segments are resumed and displayed with
the date of execution and the number of related tasks order by
status. 

645
A check-box allows to performed operation on multiple segments :
646 647 648
  - deletion : to clean unwanted data
  - tag : to tag remarkable data

649
The filter panel allows to display the segment instances with respect to 2
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
criterions :
  - tag
  - date of execution

**** The code page

Each segment names is a link to its code page. From this page the user
can view all python scripts code which have been applied to the data.

The tree view is reduced to the current segment and its related
parents.

The root path corresponding to the data storage is also displayed.


**** The product page

The number of related tasks, order by status, is a link to the product
pages, where the data can be directly displayed (if images, or text
files) or downloaded. 
From this page it is also possible to delete a specific product and
its dependencies. 


**** The log page

676
The log page can be acceded via the log button of the filter panel.
677 678 679 680
Logs are ordered by date. 



681 682
** The example pipelines
*** fft
Marc Betoule's avatar
Marc Betoule committed
683

684 685 686 687 688
**** Highlights

This example illustrates a very simple image processing use.
The problematic is the following : one wants to apply a Gaussian
filter in Fourier domain on several 2D images. 
689

690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 832 833 834 835 836 837 838 839 840 841 842 843 844 845 846 847 848 849 850 851 852
The pipe scheme is: 

#+begin_src python
pipedot = """
mkgauss->convol;
fftimg->convol;
"""
#+end_src

where segment 'mkgauss' computes the Gaussian filter, 'fftimg' computes the
Fourier transforms of the input images, and 'convol' performs the
filtering in Fourier domain, and the inverse transform of the filtered
images. 

#+begin_src python
P = pipeline.Pipeline(pipedot, code_dir=op.abspath('./'), prefix=op.abspath('./'))
P.to_dot('pipeline.dot')
#+end_src

The pipe scheme is output as a .dot file, that can be converted to an
image file with the command line: 

=dot -Tpng -o pipeline.png pipeline.dot=

To apply this filter to several images (in our case 4 input images),
the pipe data parallelism is used. 
From the main script, a 4-element list is pushed to the =fftimg=
segment. 

#+begin_src python
P.push(fftimg=[1,2,3,4]) 
#+end_src

At execution, 4 instances of the =fftimg= segment will be
created, and each of them outputs one element of this list : 

#+begin_src python
img = seg_input.values()[0] #(fftimg.py - line 16)
seg_output = [img]          #(fftimg.py - line 41)
#+end_src

On the other side, a single instance of the =mkgauss= segment will be
executed, as there is one filter to apply. 

The last segment =convol=, which depends on the two others, will be
executed with a number of instances that is the Cartesian product of
its 4+1 inputs (ie 4 instances)

The instance identifier which is set by the =fftimg= output, can be
retrieve with the following instruction: 

#+begin_src python
img = seg_input['fftimg']   #(convol.py - line 12)
#+end_src

**** Running the pipe

Follow the same procedure than for the first example pipeline, to run
this pipe and browse the result. 


*** cmb
**** Running the pipe

This CMB pipeline depends on two external python modules: 
+ healpy   :  http://code.google.com/p/healpy/
+ spherelib:  http://gitorious.org/spherelib


**** Problematic

This example illustrates a very simple CMB data processing use.  

The problematic is the following : one wants to characterize the
inverse noise weighting spectral estimator (as applied to the WMAP 1yr
data). A first demo pipeline is built to check that the algorithm
has correctly been implemented. Then, Monte Carlo simulations are used
to compute error bars estimates. 

**** A design pipeline

The design pipe scheme is: 

#+begin_src python
pipe_dot = """ 
cmb->clcmb->clplot;
noise->clcmb;
noise->clnoise->clplot;
"""
#+end_src

where: 
+ =cmb=: generate a CMB map from LCDM power spectrum. 
+ =noise=: compute the mode coupling matrix from the input hit-count map
+ =clnoise=: compute the empirical noise power spectrum from a noise
  realization. 
+ =clcmb=: generate two noise realizations, add them to the CMB map, to compute a
  first cross spectrum estimator. Then weighting mask and mode
  coupling matrix are applied to get the inverse noise weighting
  estimator
+ =clplot=: make a plot to compare pure cross spectrum vs inverse noise
  weighting estimators. 

As the two first orphan segments depends on a single shared parameter
which is the map resolution nside, this argument is pushed from the
main script. 

Another input argument of the cmb segment, is its simulation identifier,
which will be used for latter Monte Carlo. In order to push two
inputs to a single segment instance, we use python tuple data type.

#+begin_src python
P.push(cmb=[(nside, 1)])
P.push(noise=[nside])
#+end_src

From the segment, those inputs are retrieved with : 

#+begin_src python
nside  = seg_input.values()[0][0] ##(cmb.py line 13)
sim_id = seg_input.values()[0][1] ##(cmb.py line 14)
nside  = seg_input.values()[0]  ##(noise.py line 16)
#+end_src

The last segment produces a plot in which we compare: 
+ the input LCDM power spectrum
+ the binned cross spectrum of the noisy CMB maps
+ the binned cross spectrum of which we applied hitcount weight and
  mode coupling matrix. 
+ the noise power spectrum computed by clnoise segment. 

In this plot we check that both estimators are corrects, and that the
noise level is the expected one.

**** From the design pipeline to Monte Carlo

As a second step, Monte Carlo simulations are used to compute error
bars. 

The =clnoise= segment is no longer needed, so that the new pipe scheme
reads : 

#+begin_src python
pipe_dot = """ 
cmb->clcmb->clplot;
noise->clcmb;
"""
#+end_src

We now use the native data parallelization scheme of the pipe to build
many instances of the =cmb= and =clcmb= segments. 

#+begin_src python
cmbin = []
for sim_id in [1,2,3,4,5,6]:
    cmbin.append((nside, sim_id))
P.push(cmb=cmbin)
#+end_src


* Advanced usage
This section describe more complicated (and useful) features and
requires good familiarity with the concept introduced in the previous section.
853
** Multiplex directive
Marc Betoule's avatar
Marc Betoule committed
854
   
855

Marc Betoule's avatar
Marc Betoule committed
856
The default behavior can be altered by specifying a =#multiplex=
857 858 859 860 861
directive in the commentary of the segment code. If several multiplex
directives are present in the segment code the last one is retained.

The multiplex directive can be one of: 

Marc Betoule's avatar
Marc Betoule committed
862
+ =#multiplex cross_prod= : default behavior, return the Cartesian
863
  product. 
Marc Betoule's avatar
Marc Betoule committed
864
+ =#multiplex union= : make the union of the inputs
865

Marc Betoule's avatar
Marc Betoule committed
866
Moreover the =#multiplex cross_prod= directive admits filtering and
867 868
grouping by class similarly to SQL requests:

Marc Betoule's avatar
Marc Betoule committed
869
#+begin_src python
870
#multiplex cross_prod where "condition" group_by "class_function"
Marc Betoule's avatar
Marc Betoule committed
871
#+end_src
872

Marc Betoule's avatar
Marc Betoule committed
873
=condition= and =class_function= are python code evaluated for each element
874 875
of the product set. 

Marc Betoule's avatar
Marc Betoule committed
876 877
The argument of =where= is a condition. The element will be part of the
input set only if it evaluates to =True=.
878

Marc Betoule's avatar
Marc Betoule committed
879
The =group_by= directive groups elements into class according to the
880 881 882 883 884 885 886 887 888
result of the evaluation of the given class function. The input set
contains all the resulting class. For example, if the function is a
constant, the input set will contain only one element: the class
containing all elements.

During the evaluation, the values of the tuple elements are accessible
as variable wearing the name of the corresponding parents.


Marc Betoule's avatar
Marc Betoule committed
889 890 891 892 893 894 895 896
Given the Cartesian product set:
#+begin_src python
 [('Lancelot','the Brave'), ('Lancelot,'the Pure'), ('Galahad','the Brave'), ('Galahad','the
Pure')]
#+end_src

one can use :
#+begin_src python
897
#multiplex cross_prod where "quality=='the Brave'" 
Marc Betoule's avatar
Marc Betoule committed
898 899 900 901 902
#+end_src
to get 2 instances of the following segment (=melt=) running on: 
#+begin_src python
('Lancelot','the Brave'), ('Galahad','the Brave')
#+end_src
903

Marc Betoule's avatar
Marc Betoule committed
904
#+begin_src python
905
#multiplex cross_prod group_by "knights"
Marc Betoule's avatar
Marc Betoule committed
906 907 908 909 910
#+end_src
to get 2 instances of the =melt= segment running on:
#+begin_src python
('Lancelot'), ('Galahad')
#+end_src
911

Marc Betoule's avatar
Marc Betoule committed
912
#+begin_src python
913
#multiplex cross_prod group_by "0"
Marc Betoule's avatar
Marc Betoule committed
914 915
#+end_src
to get 1 instance of the =melt= segment running on: (0)
916

Marc Betoule's avatar
Marc Betoule committed
917
Note that to make use of =group_by=, elements of the output set have to be
918 919 920 921
hashable.

Another caution on the use of group: segment input data type is no
longer a dictionary in those cases as the original tuple is
Marc Betoule's avatar
Marc Betoule committed
922
lost and replaced by the result of the class function.
923

Marc Betoule's avatar
Marc Betoule committed
924
See section [[*The%20segment%20environment][The segment environment]] for more details.
925 926 927 928 929 930 931 932 933 934 935 936 937 938 939 940 941 942 943 944 945

** Depend directive

As explained in the introduction section, Pipelet offers the
possibility to spare CPU time by saving intermediate products on disk.
We call intermediate products the input/output data files of the
different segments.  

Each segment repository is identified by a unique key which depends
on: 
- the segment processing code and parameters (segment and hook
  scripts)
- the input data (identified from the key of the parent segments)

Every change made on a segment (new parameter or new parent) will then
give a different key, and tell the Pipelet engine to compute a new
segment instance.

It is possible to add some external dependencies to the key
computation using the depend directive: 

Marc Betoule's avatar
Marc Betoule committed
946
#+begin_src python
947
#depend file1 file2
Marc Betoule's avatar
Marc Betoule committed
948
#+end_src
949 950 951 952 953 954 955 956 957

At the very beginning of the pipeline execution, all dependencies will
be stored, to prevent any change (code edition) between the key
computation and actual processing.

Note that this mechanism works only for segment and hook
scripts. External dependencies are also read as the beginning of the
pipeline execution, but only used for the key computation.

Marc Betoule's avatar
Marc Betoule committed
958 959
** Database reconstruction

960 961 962
In case of unfortunate lost of the pipeline sql data base, it is
possible to reconstruct it from the disk : 

Marc Betoule's avatar
Marc Betoule committed
963
#+begin_src python
964 965
import pipelet
pipelet.utils.rebuild_db_from_disk (prefix, sqlfile)
Marc Betoule's avatar
Marc Betoule committed
966
#+end_src
Marc Betoule's avatar
Marc Betoule committed
967

968 969 970
All information will be retrieve, but with new identifiers. 

** The hooking system
971

972
As described in the 'segment environment' section, Pipelet supports
973 974 975 976 977 978
an hooking system which allows the use of generic processing code, and
code sectioning.

Let's consider a set of instructions that have to be systematically
applied at the end of a segment (post processing), one can put those
instruction in the separate script file named for example
Marc Betoule's avatar
Marc Betoule committed
979
=segname_postproc.py= and calls the hook function: 
980

Marc Betoule's avatar
Marc Betoule committed
981
#+begin_src python
982
hook('postproc', globals()) 
Marc Betoule's avatar
Marc Betoule committed
983
#+end_src
984

985
A specific dictionary can be passed to the hook script to avoid
986 987
confusion. 

Marc Betoule's avatar
Marc Betoule committed
988
The hook scripts are included into the hash key computation.
989

Marc Betoule's avatar
Marc Betoule committed
990 991
** Writing custom environments

992
The Pipelet software provides a set of default utilities available
Maude Le Jeune's avatar
Maude Le Jeune committed
993 994 995 996 997 998
from the segment environment. It is possible to extend this default
environment or even re-write a completely new one.

*** Extending the default environment

The different environment utilities are actually methods of the class
999
Environment. It is possible to add new functionalities by using the
Maude Le Jeune's avatar
Maude Le Jeune committed
1000 1001
python heritage mechanism: 

Marc Betoule's avatar
Marc Betoule committed
1002
File : =myenvironment.py=
Maude Le Jeune's avatar
Maude Le Jeune committed
1003

Marc Betoule's avatar
Marc Betoule committed
1004 1005 1006 1007 1008 1009 1010 1011 1012
#+begin_src python
  from pipelet.environment import *
  
  class MyEnvironment(Environment):
        def my_function (self):
           """ My function do nothing
           """
           return 
#+end_src
Maude Le Jeune's avatar
Maude Le Jeune committed
1013

1014
The Pipelet engine objects (segments, tasks, pipeline) are available
Marc Betoule's avatar
Marc Betoule committed
1015
from the worker attribut =self._worker=. See section [[*The%20Pipelet%20actors][The Pipelet actors]]
Marc Betoule's avatar
Marc Betoule committed
1016
for more details about the Pipelet machinery.
Maude Le Jeune's avatar
Maude Le Jeune committed
1017 1018 1019 1020 1021 1022 1023


*** Writing new environment

In order to start with a completely new environment, extend the base
environment: 

Marc Betoule's avatar
Marc Betoule committed
1024 1025
File : =myenvironment.py=
#+begin_src python
Maude Le Jeune's avatar
Maude Le Jeune committed
1026 1027 1028 1029 1030 1031 1032 1033 1034