slicing python list beyond its length - python

I have to slice a Python list/numpy array from an index to -dx and +dx.
where dx is constant. for example:
(the position/index that contains 1 only for illustration, as the center index).
A = [0, 0, 0, 0, 1, 0, 0, 0, 0]
dx=3
print(A[4-dx: 4+dx+1]) # 4 is the position of '1'
>>>[0, 0, 0, 1, 0, 0, 0]
But for this case,
B = [0, 1, 0, 0, 0 ,0, 0 ,0, 0]
print(B[1-dx: 1+dx+1])
>>>[] # because 1-dx <0.
but what i need from case B is [0, 1, 0, 0, 0]
so i did something like this, to prevent empty list/array, say n is the center index:
if n-dx <0:
result= B[:n+dx+1]
Although the above method works fine.
The original code is quite complicated, and I have to put this if...#complicated version# everywhere.
Is there any other way around? Maybe I miss something.
Thank you!

You can use the max() function to bound the index at 0.
print(A[max(0,4-dx): 4+dx+1])

Related

Python3 initializing 2D Array ignores index specification range(start,end,step)

I want to initialize an array in python3 with an index NOT starting at zero and 2 dimensions.
x_length=16
y_length=4
x_start=-4
y_start=-400
# Later passing those variables with:
level_matrix=generate_area(x_length,y_length,x_start,y_start)
Part of the function:
def generate_area(xlen,ylen,xstart,ystart):
matrix=[[0 for y in range(ystart, ystart+ylen)] for x in range(xstart, xstart+xlen)]
for index, value in enumerate(matrix):
print(f"S1_vec('{index}')<= {value}")
for index, value in enumerate(matrix[0]):
print(f"S1_vec('{index}')<= {value}")
for x in range(xstart, xstart+xlen):
for y in range(ystart, ystart+ylen):
print("Y:"+str(y))
print("X:"+str(x))
Output:
S1_vec('0')<= [0, 0, 0, 0]
S1_vec('1')<= [0, 0, 0, 0]
S1_vec('2')<= [0, 0, 0, 0]
S1_vec('3')<= [0, 0, 0, 0]
S1_vec('4')<= [0, 0, 0, 0]
S1_vec('5')<= [0, 0, 0, 0]
S1_vec('6')<= [0, 0, 0, 0]
S1_vec('7')<= [0, 0, 0, 0]
S1_vec('8')<= [0, 0, 0, 0]
S1_vec('9')<= [0, 0, 0, 0]
S1_vec('10')<= [0, 0, 0, 0]
S1_vec('11')<= [0, 0, 0, 0]
S1_vec('12')<= [0, 0, 0, 0]
S1_vec('13')<= [0, 0, 0, 0]
S1_vec('14')<= [0, 0, 0, 0]
S1_vec('15')<= [0, 0, 0, 0]
S1_vec('0')<= 0
S1_vec('1')<= 0
S1_vec('2')<= 0
S1_vec('3')<= 0
Y:-400
X:-4
IndexError: list assignment index out of range
Well, as you can clearly see, there is no negative indexes in the array. This also does not properly work with positive offset as well. The loops want to access offset index values and the script obviously fails, since the index only gets created from 0 to the var1 in case of in range(var1, var2). This makes no sense, since it should work like: range(starting_point, end_point, steps_if_needed). And the copy paste for loop syntax gets successfully used later in the script in multiple instances.
What causes such weird behavior and how to fix this without changing anything else except the initialization of the array in the code? I need 2D arrays to exactly work within the specified region. Is there a simple solution?
Edit:
Just to clarify the goal:
I need a 2D array with negative index capabilities. The range is known, but each entry needs to be defined. Append is useless, because it will not add a specific negative index for me.
If I for example need to define matrix[-4][-120]="Jeff", this needs to work. I do not even care at all, if there is a solution like in Bash, where you have to write matrix["-4,-120"]. Yet I need a reasonable way to address such entries.
You can use a virtual indexing strategy to do the same. Here is my approach:
offset = 5
size = 4
a = [x for x in range(offset,offset+size)]
print(a) # [5, 6, 7, 8]
def get_loc(idx):
return a[idx-offset]
def set_loc(idx, val):
a[idx-offset] = val
set_loc(6,15)
print(a) # [5, 15, 7, 8]
set_loc(8, 112)
print(a) # [5, 15, 7, 112]
This code was just to understand the what i mean by virtual indexing strategy, you can simply use below to get and set values:
# to get a[n] use a[n-offset]
print(a[8-offset]) # 112
# to set a[n] = x | Use a[n-offset] = x
a[7-offset] = 21
print(a) # [5, 15, 21, 112]

Counting gaps for a tetris ai in python

I am trying to make a simple tetris ai in python(no genetic algorithms)
I want to count the gaps in the grid and make the best choice depending on it.
By gap I mean where you wont be able to place a piece without clearing some lines.
My grid is something like this:
[0, 0, 0, 0, 0]
['#ff0000', ....]
[...]
0 represents a blank space, while the hex code represents its covered by a block
I have tried to calculate gaps like this:
def grid_gaps(grid):
gaps = 0
for x in range(len(grid[0])):
for y in range(len(grid)):
if grid[y][x] == 0 and \
(y > 0 and grid[y - 1][x] != 0):
gaps += 1
return gaps
It works good when the grid is like this:
[0, 0, 0, 0, 0],
[0, 0, 0, 0, 0],
[1, 1, 1, 0, 0],
[0, 0, 0, 1, 0]
1 is some color, it correctly tells me that there are 3 gaps but when the grid is someting like this:
[0, 0, 0, 0, 0],
[0, 0, 0, 0, 0],
[1, 1, 1, 0, 0],
[0, 0, 0, 1, 0],
[0, 0, 0, 1, 0]
It again returns 3 but I want it to return 6.
I think the problem is that and grid[y - 1][x] != 0 is only looking at the cell directly above the current cell, so your bottom 3 cells in the second example aren't being counted.
One quick fix I can think of is to set a gap cell to some non-zero value once it's counted, that way the gap cells below will be counted too. (Then set them back to 0 after you're done, if you're using the same grid and not a copy for the rest of the game.)
The problem is that you're looking "up" to see whether there's a blocker, but you're only looking up one row. I think you want to reorganize this so you iterate over columns, and for each column, iterate down until you hit a 1, and then continue iterating and add to the gap count for each 0 that's encountered.

Python — Randomly fill 2D array with set number of 1's

Suppose I have a 2D array (8x8) of 0's. I would like to fill this array with a predetermined number of 1's, but in a random manner. For example, suppose I want to place exactly 16 1's in the grid at random, resulting in something like this:
[[0, 0, 0, 1, 0, 0, 1, 0],
[1, 0, 0, 0, 0, 0, 0, 1],
[0, 0, 1, 1, 1, 0, 0, 0],
[0, 0, 0, 0, 0, 0, 0, 0],
[0, 1, 0, 0, 0, 0, 0, 0],
[0, 0, 1, 0, 1, 1, 0, 0],
[0, 1, 0, 0, 0, 1, 0, 0],
[0, 1, 1, 0, 0, 0, 0, 1]]
The resulting placement of the 1's does not matter in the slightest, as long as it is random (or as random as Python will allow).
My code technically works, but I imagine it's horrendously inefficient. All I'm doing is setting the probability of each number becoming a 1 to n/s, where n is the number of desired 1's and s is the size of the grid (i.e. number of elements), and then I check to see if the correct number of 1's was added. Here's the code (Python 2.7):
length = 8
numOnes = 16
while True:
board = [[(random.random() < float(numOnes)/(length**2))*1 for x in xrange(length)] for x in xrange(length)]
if sum([subarr.count(1) for subarr in board]) == 16:
break
print board
While this works, it seems like a roundabout method. Is there a better (i.e. more efficient) way of doing this? I foresee running this code many times (hundreds of thousands if not millions), so speed is a concern.
Either shuffle a list of 16 1s and 48 0s:
board = [1]*16 + 48*[0]
random.shuffle(board)
board = [board[i:i+8] for i in xrange(0, 64, 8)]
or fill the board with 0s and pick a random sample of 16 positions to put 1s in:
board = [[0]*8 for i in xrange(8)]
for pos in random.sample(xrange(64), 16):
board[pos//8][pos%8] = 1
I made the ones, made the zeros, concatenated them, shuffle them, and reshaped.
import numpy as np
def make_board(shape, ones):
o = np.ones(ones, dtype=np.int)
z = np.zeros(np.product(shape) - ones, dtype=np.int)
board = np.concatenate([o, z])
np.random.shuffle(board)
return board.reshape(shape)
make_board((8,8), 16)
Edit.
For what it's worth, user2357112's approach with numpy is fast...
def make_board(shape, ones):
size = np.product(shape)
board = np.zeros(size, dtype=np.int)
i = np.random.choice(np.arange(size), ones)
board[i] = 1
return board.reshape(shape)

Number Of Matching Elements In Two Lists

I have many sets of 2 strings. I'm trying to determine the number of matching elements in these 2 strings. The rules are if the strings share a common letter, that's a point, order does matter, but each letter in the first string can only match one of the letters in the second string. So in the strings 'aaaab', 'acccc', only 1 point is awarded because there is only one 'a' to match in the second string. Here are a few examples:
aaabb bbaaa 5
aabbb bbbaa 5
aaabb aabbb 4
aaabb ccaaa 3
aaaaa bbbbb 0
ababa babab 4
aabcc babaf 3
abcde abfgh 2
bacde abdgh 3
Hopefully that gets across how it works.
Here is the most efficient code I've been able to come up with, but its horribly convoluted. I hoping someone could think of something better.
def Score(guess, solution):
guess = list(guess)
solution = list(solution)
c = 0
for g in guess:
if g in solution and g != "_":
c += 1
solution[solution.index(g)] = "_"
return c
Surely this isn't the best way to do this, but I haven't been able to figure anything else out. I tried creating an algorithm with Counter and doing guess&solution, which worked, but ended up being way slower. Anyone have any ideas?
You could gain a ~10%* speed up by simply using the remove() method of list instead of the lookup with index().
Also, you don't need to copy guess into a list.
def Score(guess, solution):
solution = list(solution)
c = 0
for g in guess:
if g in solution:
c += 1
solution.remove(g)
return c
*at least that's what I measured on my machine
You can do it in vectorized form using NumPy!
import numpy as np
counts1 = np.bincount(np.array('aaadez', 'c').view(np.uint8), minlength=128)
counts2 = np.bincount(np.array('eeeedddddaa', 'c').view(np.uint8), minlength=128)
np.min((counts1, counts2), axis=0).sum()
counts1 looks like this:
array([0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 3, 0, 0, 1, 1, 0...])
This is an array indexed by ASCII codes. The nonzero elements are at positions 97, 100, and 101, which are ASCII 'a', 'd', and 'e'. Then we do a pairwise min(), followed by sum to get the score (in this example, 4).
Something neat about this solution is that you can apply it to as many strings as you want with no decrease in efficiency, and even very long strings will be quite fast because there are no loops in Python itself--only in compiled NumPy code.
Before editing I had a similar but slower and more complex solution using Pandas and SciPy. Here it is:
import scipy.stats
import numpy as np
import pandas
x1 = scipy.stats.itemfreq(np.array('aaade', 'c').view(np.uint8))
x2 = scipy.stats.itemfreq(np.array('bbacadde', 'c').view(np.uint8))
merged = pandas.merge(pandas.DataFrame(x1), pandas.DataFrame(x2), on=0)
np.sum(np.min(merged.values[:,1:], axis=1))
That gives 4.0. The first two lines convert the strings to arrays of integers and runs itemfreq() to count how many times each character occurs. In this example, x1 is:
arrray([[ 97., 3.],
[ 100., 1.],
[ 101., 1.]])
Then we join the two tables by the 0th column, dropping any characters that do not exist in the other one:
0 1_x 1_y
0 97 3 2
1 100 1 2
2 101 1 1
Then we just do a min and sum to get the final score (2+1+1 in this case).
Here is ;
list_a = list("aabbb")
list_b = list("bbbaa")
list_c = set(list_b)
counter = 0
for i in list_c:
if i in list_b:
counter = list_a.count(i)
print "counter : %s element : %s" %(counter,i )
I just wanna show how to count the common elements, you can change the code as summing the counter result.
here's a pretty simple solution using Counter:
def proc(vals):
for s1, s2 in vals:
c1, c2 = Counter(s1), Counter(s2)
same = set(s1) & set(s2)
print s1, s2, sum(min(c1[c], c2[c]) for c in same)
where vals looks like
vals = [('aaaaa', 'bbbbb'), ...]
Try this:
a, b = 'aaabb', 'ccaaa'
dict_a, dict_b = {}, {}
for key in list(a):
dict_a[key] = dict_a.setdefault(key, 0) + 1
for key in list(b):
dict_b[key] = dict_b.setdefault(key, 0) + 1
count = 0
for key, a_val in dict_a.items():
try:
b_val = dict_b[key]
count += min(b_val, a_val)
except KeyError:
None
print count
Same concept as #sloth, but using try instead of if
def Score(guess, solution):
solution = list(solution)
c = 0
for g in guess:
try:
solution.remove(g)
c += 1
except ValueError:
pass
return c

How to extract the bits of larger numeric Numpy data types

Numpy has a library function, np.unpackbits, which will unpack a uint8 into a bit vector of length 8. Is there a correspondingly fast way to unpack larger numeric types? E.g. uint16 or uint32. I am working on a question that involves frequent translation between numbers, for array indexing, and their bit vector representations, and the bottleneck is our pack and unpack functions.
You can do this with view and unpackbits
Input:
unpackbits(arange(2, dtype=uint16).view(uint8))
Output:
[0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0 0 0 0 0 0 0]
For a = arange(int(1e6), dtype=uint16) this is pretty fast at around 7 ms on my machine
%%timeit
unpackbits(a.view(uint8))
100 loops, best of 3: 7.03 ms per loop
As for endianness, you'll have to look at http://docs.scipy.org/doc/numpy/user/basics.byteswapping.html and apply the suggestions there depending on your needs.
This is the solution I use:
def unpackbits(x, num_bits):
if np.issubdtype(x.dtype, np.floating):
raise ValueError("numpy data type needs to be int-like")
xshape = list(x.shape)
x = x.reshape([-1, 1])
mask = 2**np.arange(num_bits, dtype=x.dtype).reshape([1, num_bits])
return (x & mask).astype(bool).astype(int).reshape(xshape + [num_bits])
This is a completely vectorized solution that works with any dimension ndarray and can unpack however many bits you want.
I have not found any function for this too, but maybe using Python's builtin struct.unpack can help make the custom function faster than shifting and anding longer uint (note that I am using uint64).
>>> import struct
>>> N = np.uint64(2 + 2**10 + 2**18 + 2**26)
>>> struct.unpack('>BBBBBBBB', N)
(2, 4, 4, 4, 0, 0, 0, 0)
The idea is to convert those to uint8, use unpackbits, concatenate the result. Or, depending on your application, it may be more convenient to use structured arrays.
There is also built-in bin() function, which produces string of 0s and 1s, but I am not sure how fast it is and it requires postprocessing too.
This works for arbitrary arrays of arbitrary uint (i.e. also for multidimensional arrays and also for numbers larger than the uint8 max value).
It cycles over the number of bits, rather than over the number of array elements, so it is reasonably fast.
def my_ManyParallel_uint2bits(in_intAr,Nbits):
''' convert (numpyarray of uint => array of Nbits bits) for many bits in parallel'''
inSize_T= in_intAr.shape
in_intAr_flat=in_intAr.flatten()
out_NbitAr= numpy.zeros((len(in_intAr_flat),Nbits))
for iBits in xrange(Nbits):
out_NbitAr[:,iBits]= (in_intAr_flat>>iBits)&1
out_NbitAr= out_NbitAr.reshape(inSize_T+(Nbits,))
return out_NbitAr
A=numpy.arange(256,261).astype('uint16')
# array([256, 257, 258, 259, 260], dtype=uint16)
B=my_ManyParallel_uint2bits(A,16).astype('uint16')
# array([[0, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0, 0, 0, 0, 0, 0],
# [1, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0, 0, 0, 0, 0, 0],
# [0, 1, 0, 0, 0, 0, 0, 0, 1, 0, 0, 0, 0, 0, 0, 0],
# [1, 1, 0, 0, 0, 0, 0, 0, 1, 0, 0, 0, 0, 0, 0, 0],
# [0, 0, 1, 0, 0, 0, 0, 0, 1, 0, 0, 0, 0, 0, 0, 0]], dtype=uint16)

Categories