#!/usr/bin/env python

# =================================================================

# File: toyprob.py

# =================================================================

#

# =================================================================

# Module: Subroutines that define the problem

# =================================================================

#

# Last update of any of the components of this module:

#

# March 25, 2008.

#

# Users are encouraged to download periodically updated versions of

# this code at the TANGO home page:

#

# www.ime.usp.br/~egbirgin/tango/

#

# ******************************************************************

# ******************************************************************

#the change in this one is going to be that we are going to include both tag bigrams and tags

#I have to change the optimization here. I will only optimze every set of conditional probabilities on its own. So, there will be 44 optimizations in each iteration

from numpy import *

import algencan

import commands

import re

from math import e

import sys

import copy

probability_re = re.compile(r'Corpus probability=(.*) per-symbol')

#implementation of the em algorithm

#import numpy

#import array

import latticeNode

#import emMethodsMDLVer2 as emMethods

import emMethodsGeneralNoPrint as emMethods

import copy

import sys

#global variables

current_optimization_params = '' #this will hold the parameters that are currently being optmized

fractional_counts_language = {}

fractional_counts_channel = {}

probabilities_channel = {}

probabilities_language = {}

current_probabilities = {}

current_fractional_counts = {}

#constraint_parameters = []

initial_parameter_args = {}

current_initial_parameter_args = {}

parameter_counter = 0

num_constraints = 1

constraint_tags_dict = {} #this will hold the tag for the constraint. the key is the constraint id and the value is the tag corresponding to this constraint

sigma = float(0)

alpha = float(0)

global_num_parameters = 0

init_option = ''

current_optimization_tag = '' #this will hold the of the constraint for which we are doing the optimization

def inip():

return n,x,l,u,m,lambda_,equatn,linear,coded,checkder

# ******************************************************************

# ******************************************************************

def evalf(x):

return -func_value,flag

# ******************************************************************

# ******************************************************************

def evalg(x):

return g,flag

# ******************************************************************

# ******************************************************************

def evalh(x):

#return ()

# ******************************************************************

# ******************************************************************

def evalc(x,ind):

return c,flag

# ******************************************************************

# ******************************************************************

def evaljac(x,ind):

return jcvar,jcval,jcnnz,flag

# ******************************************************************

# ******************************************************************

def evalhc(x,ind):

return hclin,hccol,hcval,hcnnz,flag

# ******************************************************************

# ******************************************************************

def evalfc(x,m):

return f,c,flag

# ******************************************************************

# ******************************************************************

def evalgjac(x,m):

return g,jcfun,jcvar,jcval,jcnnz,flag

# ******************************************************************

# ******************************************************************

def evalhl(x,m,lambda_,scalef,scalec):

return hllin,hlcol,hlval,hlnnz,flag

# ******************************************************************

# ******************************************************************

def evalhlp(x,m,lambda_,scalef,scalec,p,goth):

return hp,goth,flag

# ******************************************************************

# ******************************************************************

def endp(x,l,u,m,lambda_,equatn,linear):

pass

#initializing the parameter vectors. Here, each parameters is given a number which is the index in the parameter vector that algencan optimizes. The constraints are also given a number . We need these constraint numbers for algencan

#the new technique here does the optimization for each constraint separately. Therefore,it makes sense to prepare these in advance and then send them off to algencan when they are needed

#here, we will create the indexing of the parameters needed by algencan so that we can scale it. Instead of throwing the entire problem to algencan, we will #also, I will call this just once because the only code that is affected by being called in different places is the one that checks if the probability is zero. Since we will not make any probabilities zero in algencan, we just need to call this once

def createParametersForScaling(probabilities) :

internal_parameter_counter += 1

# global parameter_counter, num_constraints

# parameter_counter = internal_parameter_counter

#num_constraints = internal_num_constraints

def main() :