find weight matrix for Temporal Associative Memories - python

I am finding weight matrix for Temporal Associative Memories my vectors are
x1 = [ 1 –1 1 –1 1 –1 1 –1 ]
x2 = [ 1 1 –1 –1 1 1 –1 –1 ]
x3 = [ 1 1 1 1 –1 –1 –1 –1 ]
and equation to find weight matrix is W = T(X1)*X2 + T(X2)*X3 + T(X3)*X1 where T is transpose
x1 = [1, -1, 1, -1, 1, -1, 1, -1]
x2 = [1, 1, -1, -1, 1, 1, -1, -1]
x3 = [1, 1, 1, 1, -1, -1, -1, -1]
x = pd.DataFrame(data=[x1, x2, x3]).T
w = np.zeros(np.max(x.shape), np.max(x.shape))
for n in range(x.shape[0] - 1):
w1 = np.dot(x[n], x[n + 1])
w += w1
w += np.dot([-1], [1])
print(w)

First of all, there are brackets missing in initialization of w, it should be w = np.zeros((np.max(x.shape), np.max(x.shape))) to get proper shape.
Second, you want to iterate over columns of dataframe, so you need to use
for n in range(x.shape[1]):
Note that this will give you n values [0,1,2] not [1,2,3] (as in the book you're referring to), but this is ok since column names in pandas are by default numbered from zero as well.
In the last iteration this will give you n+1=3 and you want to address first column (indexed with 0) instead. This can be corrected with if statement.
When it comes to vector multiplication - np.dot, which you are using, performs inner product. It will treat the first vector as horizontal and second as vertical and then perform matrix multiplication. Instead, you should use np.outer, to get reversed alignment of vectors.
Last thing - I do not know what the last line inside the loop is doing
w += np.dot([-1], [1])
this does not seem to be needed (based on the book you're referring to)
Putting this all together gives following code:
x1 = [1, -1, 1, -1, 1, -1, 1, -1]
x2 = [1, 1, -1, -1, 1, 1, -1, -1]
x3 = [1, 1, 1, 1, -1, -1, -1, -1]
x = pd.DataFrame(data=[x1, x2, x3]).T
w = np.zeros((np.max(x.shape), np.max(x.shape)))
for n in range(x.shape[1]):
next_idx = n+1
if next_idx == x.shape[1]:
next_idx = 0
w1 = np.outer(x[n], x[next_idx])
w += w1
print(w)

Related

Neater strategy to drop down the numbers in the 2D list

I have a problem. It is a 2D list of non-negative integers will be given like
0, 0, 2, 0, 1
0, 2, 1, 1, 0
3, 0, 2, 1, 0
0, 0, 0, 0, 0
I have to drop the numbers, number columns. e.g. drop down the 1's down 1 column, the 2's down 2 columns, the 3's down 3 columns, and so on. If the number can't be moved down enough, wrap it around the top. (e. g If there is a 3 in the second-to-last row, it should wrap around to the first row.) If two numbers map to the same slot, the biggest number takes that slot.
After this transformation the given matrix above will end up like:
0, 0, 2, 0, 0
3, 0, 0, 0, 1
0, 0, 2, 1, 0
0, 2, 0, 1, 0
Here's my trivial solution to the problem (Assumes a list l is pre-set):
new = [[0] * len(l[0]) for _ in range(len(l))]
idx = sorted([((n + x) % len(l), m, x) for n, y in enumerate(l) for m, x in enumerate(y)], key=lambda e: e[2])
for x, y, z in idx:
new[x][y] = z
print(new)
The strategy is:
Build a list new with 0s of the shape of l
Save the new indices of each number in l and each number as tuple pairs in idx
Sort idx by each number
Assign indices from idx to the respective numbers to new list
Print new
I am not satisfied with this strategy. Is there a neater/better way to do this? I can use numpy.
Let's say you have
a = np.array([
[0,0,2,0,1],
[0,2,1,1,0],
[3,0,2,1,0],
[0,0,0,0,0]])
You can get the locations of the elements with np.where or np.nonzero:
r, c = np.nonzero(a)
And the elements themselves with the index:
v = a[r, c]
Incrementing the row is simple now:
new_r = (r + v) % a.shape[0]
To settle collisions, sort the arrays so that large values come last:
i = v.argsort()
Now you can assign to a fresh matrix of zeros directly:
result = np.zeros_like(a)
result[new_r[i], c[i]] = v[i]
The result is
[[0 0 2 0 0]
[3 0 0 0 1]
[0 0 2 1 0]
[0 2 0 1 0]]
I suggest doing it like this if only because it's more readable :-
L = [[0, 0, 2, 0, 1],
[0, 2, 1, 1, 0],
[3, 0, 2, 1, 0],
[0, 0, 0, 0, 0]]
R = len(L)
NL = [[0]*len(L[0]) for _ in range(R)]
for i, r in enumerate(L):
for j, c in enumerate(r):
_r = (c + i) % R
if c > NL[_r][j]:
NL[_r][j] = c
print(NL)

Dynamic way to compute linear constraints with multiple operators

Imagine a matrix A having one column with a lot of inequality/equality operators (≥, = ≤) and a vector b, where the number of rows in A is equal the number of elements in b. Then one row, in my setting would be computed by, e.g
dot(A[0, 1:], x) ≥ b[0]
where x is some vector, column A[,0] represents all operators and we'd know that for row 0 we were suppose to calculate using ≥ operator (e.i. A[0,0] == "≥" is true). Now, is there a way for dynamically calculate all rows in following so far imaginary way
dot(A[, 1:], x) A[, 0] b
My hope was for a dynamic evaluation of each row where we evaluate which operator is used for each row.
Example, let
A = [
[">=", -2, 1, 1],
[">=", 0, 1, 0],
["==", 0, 1, 1]
]
b = [0, 1, 1]
and x be some given vector, e.g. x = [1,1,0] we wish to compute as following
A[,1:] x A[,0] b
dot([-2, 1, 1], [1, 1, 0]) >= 0
dot([0, 1, 0], [1, 1, 0]) >= 1
dot([0, 1, 1], [1, 1, 0]) == 1
The output would be [False, True, True]
If I understand correctly, this is a way to do that operation:
import numpy as np
# Input data
a = [
[">=", -2, 1, 1],
[">=", 0, 1, 0],
["==", 0, 1, 1]
]
b = np.array([0, 1, 1])
x = np.array([1, 1, 0])
# Split in comparison and data
a0 = np.array([lst[0] for lst in a])
a1 = np.array([lst[1:] for lst in a])
# Compute dot product
c = a1 # x
# Compute comparisons
leq = c <= b
eq = c == b
geq = c >= b
# Find comparison index for each row
cmps = np.array(["<=", "==", ">="]) # This array is lex sorted
cmp_idx = np.searchsorted(cmps, a0)
# Select the right result for each row
result = np.choose(cmp_idx, [leq, eq, geq])
# Convert to numeric type if preferred
result = result.astype(np.int32)
print(result)
# [0 1 1]

How could I distribute obstacles to my grid without writing them manually?

I'm working on A star algorithm and as my code below shown the gird is written manually and I'm thinking to make a grid with 100* 100 size. So, it will be so awful to write them manually. I need to put my starting point at (0,0) location and my goal point at (99,99) location.
I'm trying to make the grid with this line below
grid1 = [[0 for i in range(100)]for j in range(100)]
But how could I assign obstacles to this grid randomly or not randomly without touching the location of starting point and goal point?
This is below my code:
from __future__ import print_function
import random
grid = [[0, 1, 0, 0, 0, 0],
[0, 1, 0, 0, 0, 0],#0 are free path whereas 1's are obstacles
[0, 1, 0, 0, 0, 0],
[0, 1, 0, 0, 1, 0],
[0, 0, 0, 0, 1, 0]]
'''
heuristic = [[9, 8, 7, 6, 5, 4],
[8, 7, 6, 5, 4, 3],
[7, 6, 5, 4, 3, 2],
[6, 5, 4, 3, 2, 1],
[5, 4, 3, 2, 1, 0]]'''
init = [0, 0]
goal = [len(grid)-1, len(grid[0])-1] #all coordinates are given in format [y,x]
cost = 1
drone_h = 60
#the cost map which pushes the path closer to the goal
heuristic = [[0 for row in range(len(grid[0]))] for col in range(len(grid))]
for i in range(len(grid)):
for j in range(len(grid[0])):
heuristic[i][j] = abs(i - goal[0]) + abs(j - goal[1])
#if grid[i][j] == 1:
#heuristic[i][j] = 99 #added extra penalty in the heuristic map
print(heuristic)
elevation = [[0 for row in range(len(grid[0]))] for col in range(len(grid))]
for i in range(len(grid)):
for j in range(len(grid[0])):
if grid[i][j] == 1:
elevation[i][j] = random.randint(1,100)
else:
elevation[i][j] = 0
#the actions we can take
delta = [[-1, 0 ], # go up
[ 0, -1], # go left
[ 1, 0 ], # go down
[ 0, 1 ]] # go right
#function to search the path
def search(grid,init,goal,cost,heuristic):
closed = [[0 for col in range(len(grid[0]))] for row in range(len(grid))]# the referrence grid
closed[init[0]][init[1]] = 1
action = [[0 for col in range(len(grid[0]))] for row in range(len(grid))]#the action grid
x = init[0]
y = init[1]
g = 0
f = g + heuristic[init[0]][init[0]] + elevation[init[0]][init[0]]
cell = [[f, g, x, y]]
found = False # flag that is set when search is complete
resign = False # flag set if we can't find expand
while not found and not resign:
if len(cell) == 0:
resign = True
return "FAIL"
else:
cell.sort()#to choose the least costliest action so as to move closer to the goal
cell.reverse()
next = cell.pop()
x = next[2]
y = next[3]
g = next[1]
f = next[0]
if x == goal[0] and y == goal[1]:
found = True
else:
for i in range(len(delta)):#to try out different valid actions
x2 = x + delta[i][0]
y2 = y + delta[i][1]
if x2 >= 0 and x2 < len(grid) and y2 >=0 and y2 < len(grid[0]):
if closed[x2][y2] == 0 and grid[x2][y2] == 0 and elevation[x2][y2] < drone_h :
g2 = g + cost
f2 = g2 + heuristic[x2][y2] + elevation[x2][y2]
cell.append([f2, g2, x2, y2])
closed[x2][y2] = 1
action[x2][y2] = i
invpath = []
x = goal[0]
y = goal[1]
invpath.append([x, y])#we get the reverse path from here
while x != init[0] or y != init[1]:
x2 = x - delta[action[x][y]][0]
y2 = y - delta[action[x][y]][1]
x = x2
y = y2
invpath.append([x, y])
path = []
for i in range(len(invpath)):
path.append(invpath[len(invpath) - 1 - i])
print("ACTION MAP")
for i in range(len(action)):
print(action[i])
return path
a = search(grid,init,goal,cost,heuristic)
for i in range(len(a)):
print(a[i])
You could assign the grid randomly and afterwards make sure both starting and end point don't contain obstacles. For neighboring fields you could just do the same as for those two.
import random
grid1 = [[random.randint(0,1) for i in range(100)]for j in range(100)]
# clear starting and end point of potential obstacles
grid1[0][0] = 0
grid1[99][99] = 0

Detecting rectangles (sub-arrays of same element value) in 2-d list

A rectangle is defined as any rectangular-shaped section of zeros within a 2-d array of 1s and 0s. Typical example:
[
[1, 1, 1, 1, 1, 1, 1, 1, 1],
[1, 1, 1, 1, 1, 1, 1, 1, 0],
[1, 1, 1, 0, 0, 0, 1, 0, 0],
[1, 0, 1, 0, 0, 0, 1, 0, 0],
[1, 0, 1, 1, 1, 1, 1, 1, 1],
[1, 0, 1, 0, 0, 1, 1, 1, 1],
[1, 1, 1, 0, 0, 1, 1, 1, 1],
[1, 1, 1, 1, 1, 1, 1, 1, 1],
]
In this example, there are three such arrays:
My goal is to determine the coordinates (outer 3 extremeties) of each array.
I start by converting the 2-d list into a numpy array:
image_as_np_array = np.array(two_d_list)
I can then get the coordinates of all the zeros thus:
np.argwhere(image_as_np_array == 0)
But this merely provides a shortcut to getting the indices by iterating over each row and calling .index(), then combining with the index of that row within the 2-d list.
I envisage now doing something like removing any element of np.argwhere() (or np.where()) where there is only a single occurrence of a 0 (effectively disregarding any row that cannot form part of a rectangle), and then trying to align contiguous coordinates, but I'm stuck with figuring out how to handle cases where any row may contain part of more than just one single rectangle (as is the case in the 3rd and 4th rows above). Is there a numpy function or functions I can leverage?
I don't know numpy, so here's a plain Python solution:
from collections import namedtuple
Rectangle = namedtuple("Rectangle", "top bottom left right")
def find_rectangles(arr):
# Deeply copy the array so that it can be modified safely
arr = [row[:] for row in arr]
rectangles = []
for top, row in enumerate(arr):
start = 0
# Look for rectangles whose top row is here
while True:
try:
left = row.index(0, start)
except ValueError:
break
# Set start to one past the last 0 in the contiguous line of 0s
try:
start = row.index(1, left)
except ValueError:
start = len(row)
right = start - 1
if ( # Width == 1
left == right or
# There are 0s above
top > 0 and not all(arr[top-1][left:right + 1])):
continue
bottom = top + 1
while (bottom < len(arr) and
# No extra zeroes on the sides
(left == 0 or arr[bottom][left-1]) and
(right == len(row) - 1 or arr[bottom][right + 1]) and
# All zeroes in the row
not any(arr[bottom][left:right + 1])):
bottom += 1
# The loop ends when bottom has gone too far, so backtrack
bottom -= 1
if ( # Height == 1
bottom == top or
# There are 0s beneath
(bottom < len(arr) - 1 and
not all(arr[bottom + 1][left:right+1]))):
continue
rectangles.append(Rectangle(top, bottom, left, right))
# Remove the rectangle so that it doesn't affect future searches
for i in range(top, bottom+1):
arr[i][left:right+1] = [1] * (right + 1 - left)
return rectangles
For the given input, the output is:
[Rectangle(top=2, bottom=3, left=3, right=5),
Rectangle(top=5, bottom=6, left=3, right=4)]
This is correct because the comments indicate that the 'rectangle' on the right is not to be counted since there is an extra 0 sticking out. I suggest you add more test cases though.
I expect it to be reasonably fast since much of the low-level iteration is done with calls to index and any, so there's decent usage of C code even without the help of numpy.
I have written a simple algorithms using the Sweep line method. The idea is that You go through the columns of You array column by column, and detect the series of zeros as potentially new rectangles. In each column You have to check if the rectangles detected earlier have ended, and if yes add them to the results.
import numpy as np
from sets import Set
from collections import namedtuple
example = np.array([
[1, 1, 1, 1, 1, 1, 1, 1, 1],
[1, 1, 1, 1, 1, 1, 1, 1, 0],
[1, 1, 1, 0, 0, 0, 1, 0, 0],
[1, 0, 1, 0, 0, 0, 1, 0, 0],
[1, 0, 1, 1, 1, 1, 1, 1, 1],
[1, 0, 1, 0, 0, 1, 1, 1, 1],
[1, 1, 1, 0, 0, 1, 1, 1, 1],
[1, 1, 1, 1, 1, 1, 1, 1, 1],
])
Rectangle = namedtuple("Rectangle", "left top bottom right")
def sweep(A):
height = A.shape[0]
length = A.shape[1]
rectangles = dict() # detected rectangles {(rowstart, rowend): col}
result = []
# sweep the matrix column by column
for i in xrange(length):
column = A[:, i]
# for currently detected rectangles check if we should extend them or end
for r in rectangles.keys():
# detect non rectangles shapes like requesten in question edit and del those rectangles
if all([x == 0 for x in column[r[0]:r[1]+1]]) and ((r[0]-1>0 and column[r[0]-1]==0) or (r[1]+1<height and column[r[1]+1]==0)):
del rectangles[r]
elif any([x == 0 for x in column[r[0]:r[1]+1]]) and not all([x == 0 for x in column[r[0]:r[1]+1]]):
del rectangles[r]
# special case in the last column - add detected rectangles
elif i == length - 1 and all([x == 0 for x in column[r[0]:r[1]+1]]):
result.append(Rectangle(rectangles[r], r[0], r[1], i))
# if detected rectangle is not extended - add to result and del from list
elif all([x == 1 for x in column[r[0]:r[1]+1]]):
result.append(Rectangle(rectangles[r], r[0], r[1], i-1))
del rectangles[r]
newRectangle = False
start = 0
# go through the column and check if any new rectangles appear
for j in xrange(height):
# new rectangle in column detected
if column[j] == 0 and not newRectangle and j+1 < height and column[j+1] == 0:
start = j
newRectangle = True
# new rectangle in column ends
elif column[j] == 1 and newRectangle:
# check if new detected rectangle is already on the list
if not (start, j-1) in rectangles:
rectangles[(start, j-1)] = i
newRectangle = False
# delete single column rectangles
resultWithout1ColumnRectangles = []
for r in result:
if r[0] != r[3]:
resultWithout1ColumnRectangles.append(r)
return resultWithout1ColumnRectangles
print example
print sweep(example)
returns:
[[1 1 1 1 1 1 1 1 1]
[1 1 1 1 1 1 1 1 0]
[1 1 1 0 0 0 1 0 0]
[1 0 1 0 0 0 1 0 0]
[1 0 1 1 1 1 1 1 1]
[1 0 1 0 0 1 1 1 1]
[1 1 1 0 0 1 1 1 1]
[1 1 1 1 1 1 1 1 1]]
[Rectangle(left=3, top=5, bottom=6, right=4),
Rectangle(left=3, top=2, bottom=3, right=5)]

NumPy: How to avoid this loop?

Is there a way to avoid this loop so optimize the code?
import numpy as np
cLoss = 0
dist_ = np.array([0,1,0,1,1,0,0,1,1,0]) # just an example, longer in reality
TLabels = np.array([-1,1,1,1,1,-1,-1,1,-1,-1]) # just an example, longer in reality
t = float(dist_.size)
for i in range(len(dist_)):
labels = TLabels[dist_ == dist_[i]]
cLoss+= 1 - TLabels[i]*(1. * np.sum(labels)/t)
print cLoss
Note: dist_ and TLabels are both numpy arrays with the same shape (t,1)
I am not sure what you exactly want to do, but are you aware of scipy.ndimage.measurements for computing on arrays with labels? It look like you want something like:
cLoss = len(dist_) - sum(TLabels * scipy.ndimage.measurements.sum(TLabels,dist_,dist_) / len(dist_))
I first wonder, what is labels at each step in the loop?
With dist_ = array([2,1,2]) and TLabels=array([1,2,3])
I get
[-1 1]
[1]
[-1 1]
The different length immediately raise a warning flag - it may be difficult to vectorize this.
With the longer arrays in the edited example
[-1 1 -1 -1 -1]
[ 1 1 1 1 -1]
[-1 1 -1 -1 -1]
[ 1 1 1 1 -1]
[ 1 1 1 1 -1]
[-1 1 -1 -1 -1]
[-1 1 -1 -1 -1]
[ 1 1 1 1 -1]
[ 1 1 1 1 -1]
[-1 1 -1 -1 -1]
The labels vectors are all the same length. Is that normal, or just a coincidence of values?
Drop a couple of elements off of dist_, and labels are:
In [375]: for i in range(len(dist_)):
labels = TLabels[dist_ == dist_[i]]
v = (1.*np.sum(labels)/t); v1 = 1-TLabels[i]*v
print(labels, v, TLabels[i], v1)
cLoss += v1
.....:
(array([-1, 1, -1, -1]), -0.25, -1, 0.75)
(array([1, 1, 1, 1]), 0.5, 1, 0.5)
(array([-1, 1, -1, -1]), -0.25, 1, 1.25)
(array([1, 1, 1, 1]), 0.5, 1, 0.5)
(array([1, 1, 1, 1]), 0.5, 1, 0.5)
(array([-1, 1, -1, -1]), -0.25, -1, 0.75)
(array([-1, 1, -1, -1]), -0.25, -1, 0.75)
(array([1, 1, 1, 1]), 0.5, 1, 0.5)
Again different lengths of labels, but really only a few calculations. There is 1 v value for each different dist_ value.
Without working out all the details, it looks like you are just calculating labels*labels for each distinct dist_ value, and then summing those.
This looks like a groupBy problem. You want to divide the dist_ into groups with a common value, and sum some function of their corresponding TLabels values. Python itertools has a groupBy function, so does pandas. I think both require you to sort dist_.
Try sorting dist_ and see if that adds any clarity to the problem.
I'm not sure if this is any better since I didn't exactly understand why you might want to do this. Many variables in your loop are bivalued hence can be computed in advance.
Also the entries of dist_ can be used as a boolean switch but I used an explicit copy anyhow.
dist_ = np.array([0,1,0,1,1,0,0,1,1,0])
TLabels = np.array([-1,1,1,1,1,-1,-1,1,-1,-1])
t = len(dist)
dist_zeros = dist_== 0
one_zero_sum = [sum(TLabels[dist_zeros])/t , sum(TLabels[~dist_zeros])/t]
cLoss = sum([1-x*one_zero_sum[dist_[y]] for y,x in enumerate(TLabels)])
which results in cLoss = 8.2. I am using Python3 so didn't check whether this is a true division or not in Python2.

Categories