krrepo / lea Goto Github PK

Python 3 support is good and easy

It would be good to include a test suite, to validate the different functions 
of Lea. A lot of consistency checks can be made internally, between results of 
functions themselves: verify basic arithmetics, logical operators,.... 
Examples: 2*die == die+die,  (die==1) | (die>1). For more advanced tests, 
different examples with expected results could be found in the literature.
This would allow for better confidence in the tool from the users, as well as 
non-regression tests for maintenance.

Consider the following CPT example in Leapp;

x = ?:(1/3)
y = ?:(1/4)
z1 = ?! ( ~x & ~y -> ?:(1/5),
          ~x &  y -> ?:(1/7),
           x & ~y -> ?:(1/2),
           x &  y -> ?:(1/2))

Notice that z1 is independent of y when x is true (i.e. "context-specific 
Independence"). 
Then, in the following refactoring

z0 = ?! ( ~y -> ?:(1/5),
           y -> ?:(1/7))
z3 = ?! ( ~x -> z0,
          x -> ?:(1/2))

z3 is expected to be equivalent to z1.

However the two distributions unexpectedly differ:

lea> :@ z1 ! y
11/42
lea> :@ z3 ! y
8/21

Hi, here's a suggestion for a nice example using Grime's non-transitive dice, 
maybe you want to include something similar in the README? 

http://grime.s3-website-eu-west-1.amazonaws.com/

The idea behind Grime's dice is that you have three dice, red, blue, olive 
which win against each other in a cyclic way. (one beating the other in about 
5:7). When you have two dice of the same colour, they beat each other, too, but 
the other way 'round.

red   = Lea.fromVals(3,3,3,3,3,6)
blue  = Lea.fromVals(2,2,2,5,5,5)
olive = Lea.fromVals(1,4,4,4,4,4)

print(red > blue)
print(blue > olive)
print(olive > red)

print(red+red.clone() > blue+blue.clone())
...

Google Code is closing soon 
[http://google-opensource.blogspot.fr/2015/03/farewell-to-google-code.html]. Is 
there any plan to migrate lea somewhere else before that happens?

Thanks

CPT do not allow for conditions that are certainly unfeasible. Examples:

lea> a = ?('x','y')
lea> ?! (a==a -> 's',
 ...      _   -> 't')
Lea error: forbidden to define 'other' clause for complete clause set
lea> ?! (a!=a -> 's',
 ...      _   -> 't')
Lea error: some clause condition is not feasible

Even if such conditions are not usual, it should be more sensible to accept 
them and produce expected results (e.g. 's' with probability 1 in first case, 
't' with probability 1 in second case). See treatment of if-then-else in 
programming languages.

Consider the following CPT example in Leapp;

x = ?:(1/3)
y = ?:(1/4)
z1 = ?! ( ~x & ~y -> ?:(1/5),
          ~x &  y -> ?:(1/7),
           x & ~y -> ?:(1/2),
           x &  y -> ?:(1/2))

Notice that z1 is independent of y when x is true (i.e. "context-specific 
Independence"). 
Then, It seems natural to factor out the two last entries as

z2 = ?! ( ~x & ~y -> ?:(1/5),
          ~x &  y -> ?:(1/7),
           x      -> ?:(1/2))

However the two distributions unexpectedly differ:

lea> :@ z1
61/210
lea> :@ z2
139/630

z1 is right while z2 is wrong.

What steps will reproduce the problem?
1. import lea

What is the expected output? What do you see instead?

I don't expect, need, or want any output if the import goes well, instead I see 
this:

 """
 Lea 1.3.1 Copyright (C) 2013  Pierre Denis
 This program comes with ABSOLUTELY NO WARRANTY.
 This is free software, and you are welcome to redistribute it under certain conditions.
 For more details, see COPYING and COPYING.LESSER files.
 """

What version of the product are you using? On what operating system?

Lea 1.3.1 on Python 3.4.2 on Linux 3.17.2.

Please provide any additional information below.

Importing a package shouldn't spew out a banner.  Doing so makes it useless in 
scripts.

What steps will reproduce the problem?
1. Testing the withProb method as described in the [Advanced Tutorial part 
1](https://code.google.com/p/lea/wiki/LeaPyTutorial1#Revised_Distributions)

die1 = Lea.fromVals(1,2,3,4,5,6)
dieU = die1.withProb(die1 >=5, 1, 2)
print(dieU)

What is the expected output? What do you see instead?

Expected:
1 : 1/8
2 : 1/8
3 : 1/8
4 : 1/8
5 : 2/8
6 : 2/8

Seen:
1 : 1/6
2 : 1/6
3 : 1/6
4 : 1/6
5 : 1/6
6 : 1/6

What version of the product are you using? On what operating system?
2.1.1 with python 2.7.6 on Mac.

Please provide any additional information below.

Do you plan replace repo to github?

The Lea code contains several loops on dictionaries that use items() or 
values() methods. OK, this allows for the same implementation on Python 2 and 
3, which is handy. However, this is inefficient on Python 2 because these 
methods build lists instead of iterators; there are penalties on speed and 
memory. In Python 2, the right methods to call are iteritems() and 
itervalues(). Note that these methods were present in the fist versions of Lea, 
before porting to Python 3. It would be nice if Lea uses the most efficient 
methods depending on the Python version on which it runs.

Consider the following Lea session:
{{{
>>> (die1.times(2)) + die1
 3 :  1/216
 4 :  3/216
 5 :  6/216
 6 : 10/216
 7 : 15/216
 8 : 21/216
 9 : 25/216
10 : 27/216
11 : 27/216
12 : 25/216
13 : 21/216
14 : 15/216
15 : 10/216
16 :  6/216
17 :  3/216
18 :  1/216
>>> die1 + (die1.times(2))
 3 : 1/6
 6 : 1/6
 9 : 1/6
12 : 1/6
15 : 1/6
18 : 1/6
}}}
Both results should be the same! The second result can be explained by the 
binding mechanism.
A workaround is to clone _die1_ before doing the _times(2)_:
{{{
>>> die1 + (die1.clone().times(2))
 3 :  1/216
 4 :  3/216
 5 :  6/216
 6 : 10/216
 7 : 15/216
 8 : 21/216
 9 : 25/216
10 : 27/216
11 : 27/216
12 : 25/216
}}}

Resolution: change the _times_ method to perform a clone of self.

The example shown in the tutorial

die1.withProb(die1>=5,1,2)

causes the message

Lea error: unfeasible: probability shall remain 0

Consider the following CPT example in Leapp;

x = ?:(1/3)
y = ?:(1/4)
z1 = ?! ( ~x & ~y -> ?:(1/5),
          ~x &  y -> ?:(1/7),
           x & ~y -> ?:(1/2),
           x &  y -> ?:(1/2))

Notice that z1 is independent of y when x is true (i.e. "context-specific 
Independence"). 
The following refactoring

z4 = ?! ( ~x -> ?! ( ~y -> ?:(1/5),
                      y -> ?:(1/7)),
           x -> ?:(1/2))

is not correctly parsed by Leapp, which reports

Python error: too many values to unpack (expected 2)

After activating the debug mode of Leapp, the following trace is displayed:

Traceback (most recent call last):
  File "/Users/admin/Documents/lea2/leapp_console.py", line 108, in startCmdLoop
    self.execLeappTranslatorMultilineStatement(rMultilineStatement)
  File "/Users/admin/Documents/lea2/leapp_console.py", line 71, in execLeappTranslatorMultilineStatement
    pMultilineStatement = LeappTranslator.getTarget00(rMultilineStatement)
  File "/Users/admin/Documents/lea2/leapp_translator.py", line 107, in getTarget00
    pStatementLines.append(identStr+LeappTranslator.getTarget0(unindentedLeaStatementLine))
  File "/Users/admin/Documents/lea2/leapp_translator.py", line 143, in getTarget0
    return headTarget + LeappTranslator.getTarget0b(sourceFragment) + tailTarget
  File "/Users/admin/Documents/lea2/leapp_translator.py", line 151, in getTarget0b
    target = LeappTranslator.getTarget1(sourceFragment)
  File "/Users/admin/Documents/lea2/leapp_translator.py", line 189, in getTarget1
    unstringedTargetFragment = LeappTranslator.getTarget(unstringedFragment)
  File "/Users/admin/Documents/lea2/leapp_translator.py", line 266, in getTarget
    targetFragment = LeappTranslator.parse(targetFragment,'?!' ,'(',')','Lea.buildCPT(*(%s,))',LeappTranslator.treatCPTExpression)
  File "/Users/admin/Documents/lea2/leapp_translator.py", line 364, in parse
    body = treatFunc(body)
  File "/Users/admin/Documents/lea2/leapp_translator.py", line 258, in treatCPTExpression
    newCPTExpression = ','.join('(%s,%s)'%(f(condExpr),distribExpr) for (condExpr,distribExpr) in cptItems)
  File "/Users/admin/Documents/lea2/leapp_translator.py", line 258, in <genexpr>
    newCPTExpression = ','.join('(%s,%s)'%(f(condExpr),distribExpr) for (condExpr,distribExpr) in cptItems)
ValueError: too many values to unpack (expected 2)

One feature I miss is create a distribution from a list.