Browse Source

Designed a dynamic programming technique for the paragraph formatter.

pull/18/head
jrtechs 6 years ago
parent
commit
3fbc19d8d5
1 changed files with 79 additions and 15 deletions
  1. +79
    -15
      other/paragraphFormatting.py

+ 79
- 15
other/paragraphFormatting.py View File

@ -1,3 +1,10 @@
import math
"""
Jeffery Russell
11-13-18
"""
def extraSpace(S, M, i, j):
"""
@ -28,7 +35,7 @@ def badnessLine(S, M, i, j):
"""
es = extraSpace(S, M, i, j)
if es < 0:
return float("infinity")
return math.inf
return es
@ -44,19 +51,17 @@ def minBad(S, M, i):
:param M: Max length of line
:param i: start word index
"""
if len(S) == 0:
if badnessLine(S, M, i, len(S) -1) != math.inf:
return 0
# Calculate the current line's badness
curBad = 0
k = i
while curBad > badnessLine(S, M, i, k):
curBad = badnessLine(S, M, i, k)
k = k + 1
return max(curBad, badnessLine(S, M, k))
min = math.inf
for k in range(i + 1, len(S)):
end = minBad(S, M, k)
front = badnessLine(S, M, i, k - 1)
max = end if end > front else front
min = min if min < max else max
return min
def minBadDynamic(S, M):
@ -68,20 +73,63 @@ def minBadDynamic(S, M):
:param S: List of words
:param M: Max length of line
"""
pass
cost = [math.inf for i in range(len(S))]
for i in range(0, len(S)):
if badnessLine(S, M, 0, i) != math.inf:
cost[i] = badnessLine(S, M, 0, i)
if i == len(S) -1:
return 0 #Everything fit on one line
else:
min = math.inf
for k in range(0, i):
before = cost[k]
after = badnessLine(S, M, k + 1, i)
if i == len(S) -1 and badnessLine(S, M, k+1, i) != math.inf:
after = 0 # Last Line
max = before if before > after else after
min = min if min < max else max
cost[i] = min
return cost[len(S) -1]
def minBadDynamicChoice(S, M):
"""
Write a procedure minBadDynamicChoice that implements
the function mb 0 using dynamic
the function mb' using dynamic
programming. In addition to returning mb(S, M ), it
should also return the choices made
:param S: List of words
:param M: Max length of line
"""
pass
cost = [math.inf for i in range(len(S))]
# List of tuples indicating start/end indices of line
choice = [[] for i in range(len(S))]
for i in range(0, len(S)):
if badnessLine(S, M, 0, i) != math.inf:
cost[i] = badnessLine(S, M, 0, i)
choice[i] = [(0, i)]
if i == len(S) - 1:
return 0, [(0,i)] # One line
else:
min = math.inf
choiceCanidate = []
for k in range(0, i): # Finds the optimal solution
before = cost[k] # Previously computed minimum
after = badnessLine(S, M, k + 1, i) # Badness of new slice
if i == len(S) - 1 and badnessLine(S, M, k+1, i) != math.inf:
after = 0 # Last line
max = before if before > after else after
if min > max:
# Captures where slice is being taken
choiceCanidate = choice[k] + [(k+1, i)]
min = max
choice[i] = choiceCanidate
cost[i] = min
return cost[len(S) -1], choice[len(S) -1]
def printParagraph(S, M):
@ -92,7 +140,23 @@ def printParagraph(S, M):
What is the asymptotic running time of your
algorithm?
This program will run asymptotically in O(n^2) due
to the characteristic of calculating the minBad
for each sub sequence and then looping through a
maximum of |S| to find the optimal solution.
:param S: List of words
:param M: Max length of line
"""
pass
cost, choice = minBadDynamicChoice(S, M)
print()
for i in range(0, len(choice)):
for x in range(choice[i][0], choice[i][1] + 1):
print(str(S[x]), end=" ")
print()
print(minBadDynamicChoice(["aaa", "aaa"], 6))
printParagraph(["jjjjjj","aaa", "bb", "cc", "ff", "mm", "ddddd", "ddddd"], 6)
printParagraph(["jjjjjj"], 6)

Loading…
Cancel
Save