Tom, some additional comments are in line Rick
Tom Parris wrote: > Rick, > > Thanks! This is very helpful. I'll read the Ptolemy document more > thoroughly and try again. > > The basic challenge I'm having is translating from java examples to > python. I've been experimenting with the optional methods. In the > Java examples I've been looking at (e.g., Repeat, DownSample) the > constructor method is used to: create and set parameters, and sets the > input_tokenConsumptionRate and output_tokenConsumptionRate. The same > is true for SDFTranformer (from which these examples are subclassed). > That is what led me to try the same in Python (with the __init__ > method). > the initialize(0 and preinitialize() methods provide support for what you would normally do in __init__(). preinitialize() is called just once in each run, however, there are circumstances where initialize() may be called more than once ... I don;t understand what those circumstances are. In practice, I use preinitialize() to read true parameters (those defined in "Configure Actor") and store them into Python instance attributes. > The other challenge I'm having is reading n elements from an input > port, but only consuming m elements (where 0<=m<=n) in order to > implement sliding window actors. Given the code for IOPort.java, I'm > not convinced this is even possible. It looks (to the untrained eye) > that every call to get implicitly pops tokens off the channel. You must read all tokens from a port on each fire. My suggestion is to read the all of the tokens and either store the extras in a Python attribute or just ignore them. > > I'll work both challenges some more and get back to you offline if I > still need help. > > -- Tom > > On Tue, Nov 24, 2009 at 11:04 AM, Rick Moore <rem63 at cornell.edu > <mailto:rem63 at cornell.edu>> wrote: > > Tom, > > I ran your workflow and both ports had the correct values ... but > your script doesn't currently do anything with the blocksize > except print it to stdout (which in kepler is the console). And > that works as well. > > A PythonActor doesn't communicate with it's scripts through the > standard Python __init__ method. Rather it expects the Python > scripts to implement the same methods that a Java actor might > implement. Of course, you already use fire(), but there are > several other optional methods available : > > def preinitialize(self) ... set port types and/or scheduling > information ... > def initialize(self) ... initialize local variables ... > def prefire(self) ... determine whether firing should proceed ... > a good place to check whether the execution environment is valid > if you use it, it must return True to proceed, False if > there is some problem ... no return value is the same as False > the fire(0 method will be called only if prefire() returns True > def postfire(self) ... update persistent state and/or determine > whether firing should continue to next iteration > if you use it, it must return True to proceed, False if > there is some problem ... no return value is the same as False > returning False will terminate the workflow > def wrapup(self) ... cleanup and display final results ... > I use this to reset to the initial state after the last iteration > > Designing actors is discussed in detail in chapter 5 of the > document at > http://www.eecs.berkeley.edu/Pubs/TechRpts/2008/EECS-2008-28.pdf > > If you want, I can help you in more detail off list. > > Rick Moore > Information Science > Cornell University > > "Once in a while you will be shown the light in the strangest of > places, if you look at it right" > from Scarlet Begonias by the Grateful Dead > > Tom Parris wrote: >> Dear Keplerites: >> >> I have been making steady progress understanding how to write >> actors for Kepler using the PythonScript actor. However, I have >> hit a stumbling block in my attempts to set >> input_tokenConsumptionRate. >> >> To illustrate my problem, I have attached a simple workflow with >> an actor titled "IntTokenSum" written with the PythonScript >> actor. This actor is designed to add some number of integer >> tokens from it's "input" port. The number of tokens is defined >> by the "windowsize" port. The number of tokens to consume with >> each "fire" is defined with the "blocksize" port. The current >> version works as desired with the exception of the blocksize >> input prot (it always takes on the value of the windowsize input >> port). >> >> Q1: As the script is currently configured, the windowsize >> parameter works as advertised. However, I can't seem to get the >> blocksize parameter to work. I've tried a number of approaches >> to setting the input_tokenConsumptionRate per the java code for >> several actors (e.g., DownSample). But now I'm wondering if that >> is the correct approach. Perhaps its the call to self.input.get >> that needs to be changed. Correct output on Display2 should be >> the sequence: 1, 3, 5, 7, 9, 11, 13, 15, 17. >> >> Q2: I'd also like to use an actor along these lines under the SDF >> director with portparameters for windowsize and blocksize. But >> can't figure out how to to this with the PythonScript actor. Is >> this possible? If so, I would welcome guidance. >> >> Q3: Finally, it should be fairly easy to make this actor >> polymorphic. I think this can be done by changing the line: >> >> val = val + it.intValue() >> >> to >> >> val = it.add(val) >> >> But I am not sure how to initialize val to the additive identity >> (generalized notion of zero). >> >> ------------------------------------------------------------------------ >> >> _______________________________________________ >> Kepler-users mailing list >> Kepler-users at kepler-project.org <mailto:Kepler-users at >> kepler-project.org> >> http://mercury.nceas.ucsb.edu/kepler/mailman/listinfo/kepler-user > -------------- next part -------------- An HTML attachment was scrubbed... URL: <http://mercury.nceas.ucsb.edu/kepler/pipermail/kepler-users/attachments/20091124/3c9def07/attachment-0001.html>
