Custom tetris block placement algorithm - air
Well it's not entirely a generic problem, but that's why we're here after all.
A bit of background
My task is to create a tetris like simulation with an AI playing it. The lines do not disappear when they are completed. The end result should be a matrix filled with the neatly placed blocks, with very few or no gaps.
What I chose to do, was a genetic approach, with constant weights and methods for evaluation. The AI would try to place the blocks in all possible places, rotations, evaluate the temporary matrices, and go with the best one.
The problem
In tetris, you can move to the left or right even when the block is on the ground. This allows to solve many positions that would otherwise be impossible. The real problem however, that these holes can even occur mid-air, something like this:
falling J shape, with optimal choice occurring mid-air
The only solution I see here, would be trying all positions, rotations, and all possible combinations of mid-air moves, which I assume is "not an optimal solution" to say it formally.
My question
Is if someone has an idea or another approach, to find these possibilities for placement with realistic amounts of computing power
A single piece can be positioned in a maximum of 10x20x4 = 800 positions on a single board. These will be the nodes of your graph. Where you can move from one node to another in a single move, there is an edge. You can then prune nodes that are illegal (e.g. overlap with existing obstacles on the board). You can also mark nodes as "final" - there, the tile has an obstacle directly under at least one part of it (but they can still have nodes connected to them).
You can then check which final nodes are reachable from the initial node, which is a standard problem.
You could optimize this further by ignoring nodes where the piece is above the current height of the board.
Example code:
import copy
import time
from os import system, name
from random import randint
from wrapt import synchronized
from asciimatics.screen import Screen
import asciimatics
from asciimatics.effects import Cycle, Stars
from asciimatics.renderers import FigletText
from asciimatics.scene import Scene
from asciimatics.screen import Screen
class Tile:
shapes = \
[
[
[0, 0, 2],
[2, 2, 2]
],
[
[3, 3, 0],
[0, 3, 3]
],
[
[0, 4, 4],
[4, 4, 0]
],
[
[5, 0, 0],
[5, 5, 5]
],
[
[6, 6],
[6, 6]
],
[
[0, 0, 0, 0],
[7, 7, 7, 7],
[0, 0, 0, 0]
],
[
[0, 8, 0],
[8, 8, 8]
]
]
def __init__(self, id=-1):
if id >= 0:
self.id = id
else:
self.id = randint(0, len(self.shapes)-1)
self.shape = self.shapes[id]
x = 8
y = 0
id = 0
def rotate(self):
self.shape = list(zip(*self.shape[::-1]))
class Model:
_height = 25
_width = 20
_score = 0
def __init__(self):
self._view = None
self._field = [[0] * self._width for i in range(self._height)]
for i in range(5):
for j in range(self._height):
self._field[j][i] = 1
self._field[j][-i-1] = 1
for i in range(5):
for j in range(self._width):
self._field[-i-1][j] = 1
self._tile = Tile()
self._nexttile = Tile()
def set_view(self, view):
self._view = view
def get_height(self):
i = 0
for r in self._field[:-5]:
full_line = True
if sum(r[5:-5]) > 0:
return i
i += 1
return i
def _merge(self, field, tile):
field_copy = copy.deepcopy(field)
for i in range(len(tile.shape)):
for j in range(len(tile.shape[0])):
field_copy[tile.y + i][tile.x + j] += tile.shape[i][j]
return field_copy
#synchronized
def _is_valid(self, field, tile):
for i in range(len(tile.shape)):
for j in range(len(tile.shape[0])):
if tile.shape[i][j] > 0:
if (field[tile.y + i][tile.x + j] > 0):
return False
return True
def get_board(self):
return self._merge(self._field, self._tile)
def rotate(self):
self._tile.rotate()
if not self._is_valid(self._field, self._tile):
self._tile.rotate()
self._tile.rotate()
self._tile.rotate()
if self._view is not None:
self._view.show()
def _which_lines_completed(self):
lines = []
i = 0
for r in self._field[:-5]:
full_line = True
for c in r:
if c == 0:
full_line = False
if full_line:
lines.append(i)
i += 1
return lines
def _remove_lines(self, lines):
for l in lines:
for i in list(range(1, l+1))[::-1]:
self._field[i] = self._field[i-1].copy()
if len(lines) == 4:
self._score += 5000
elif len(lines) == 3:
self._score += 1000
elif len(lines) == 2:
self._score += 500
elif len(lines) == 1:
self._score += 100
#synchronized
def move_down(self):
self._tile.y += 1
if not self._is_valid(self._field, self._tile):
self._tile.y -= 1
self._field = self._merge(self._field, self._tile)
self._tile = self._nexttile
self._nexttile = Tile()
# Check if any lines need to be removed
lines = self._which_lines_completed()
# If lines need to be removed, notify the view
if len(lines) > 0:
self._view.remove_lines(lines)
# Remove the lines
self._remove_lines(lines)
if self._view is not None:
self._view.show()
#synchronized
def move_left(self):
self._tile.x -= 1
if not self._is_valid(self._field, self._tile):
self._tile.x += 1
else:
if self._view is not None:
self._view.show()
#synchronized
def move_right(self):
self._tile.x += 1
if not self._is_valid(self._field, self._tile):
self._tile.x -= 1
if self._view is not None:
self._view.show()
class AsciimaticView:
def __init__(self, model):
self.screen = Screen.open()
self.model = model
def _show_board(self, board):
#self.screen.clear()
b = board
x = 0
y = 0
for r in b[:-4]:
x = 0
for c in r[4:-4]:
if c == 1:
self.screen.print_at(u'██', x, y, Screen.COLOUR_BLUE, Screen.A_BOLD)
elif c == 2:
self.screen.print_at(u'[]', x, y, Screen.COLOUR_RED, Screen.A_BOLD)
elif c == 3:
self.screen.print_at(u'[]', x, y, Screen.COLOUR_GREEN, Screen.A_BOLD)
elif c == 4:
self.screen.print_at(u'[]', x, y, Screen.COLOUR_GREEN, Screen.A_BOLD)
elif c == 5:
self.screen.print_at(u'[]', x, y, Screen.COLOUR_RED, Screen.A_BOLD)
elif c == 6:
self.screen.print_at(u'[]', x, y, Screen.COLOUR_CYAN, Screen.A_BOLD)
elif c == 7:
self.screen.print_at(u'[]', x, y, Screen.COLOUR_YELLOW, Screen.A_BOLD)
elif c == 8:
self.screen.print_at(u'[]', x, y, Screen.COLOUR_MAGENTA, Screen.A_BOLD)
else:
self.screen.print_at(u' ', x, y, Screen.COLOUR_BLUE, Screen.A_BOLD)
x += 2
y += 1
self.screen.print_at(u' ', 0, y, Screen.COLOUR_RED, Screen.A_BOLD)
self.screen.print_at(u' ', 0, y+1, Screen.COLOUR_RED, Screen.A_BOLD)
self.screen.print_at(u' ', 0, y+2, Screen.COLOUR_RED, Screen.A_BOLD)
self.screen.print_at(u' ', 0, y+3, Screen.COLOUR_RED, Screen.A_BOLD)
for i in range(len(self.model._nexttile.shape)):
x = 0
if (i == 1):
self.screen.print_at(u'Next: ', x, y, Screen.COLOUR_WHITE, Screen.A_BOLD)
x = x + 6
for j in range(len(self.model._nexttile.shape[0])):
c = self.model._nexttile.shape[i][j]
if c == 1:
self.screen.print_at(u'██', x, y, Screen.COLOUR_BLUE, Screen.A_BOLD)
elif c == 2:
self.screen.print_at(u'[]', x, y, Screen.COLOUR_RED, Screen.A_BOLD)
elif c == 3:
self.screen.print_at(u'[]', x, y, Screen.COLOUR_GREEN, Screen.A_BOLD)
elif c == 4:
self.screen.print_at(u'[]', x, y, Screen.COLOUR_GREEN, Screen.A_BOLD)
elif c == 5:
self.screen.print_at(u'[]', x, y, Screen.COLOUR_RED, Screen.A_BOLD)
elif c == 6:
self.screen.print_at(u'[]', x, y, Screen.COLOUR_CYAN, Screen.A_BOLD)
elif c == 7:
self.screen.print_at(u'[]', x, y, Screen.COLOUR_YELLOW, Screen.A_BOLD)
elif c == 8:
self.screen.print_at(u'[]', x, y, Screen.COLOUR_MAGENTA, Screen.A_BOLD)
else:
self.screen.print_at(u' ', x, y, Screen.COLOUR_BLUE, Screen.A_BOLD)
x = x + 2
y = y + 1
x = 0
y = 24
self.screen.print_at(u'Score: ' + str(self.model._score), x, y, Screen.COLOUR_WHITE, Screen.A_BOLD)
self.screen.refresh()
def show(self):
self._show_board(self.model.get_board())
def remove_lines(self, lines):
b = self.model.get_board()
for i in range(5):
for l in lines:
b[l][5:-5] = [1-el for el in b[l][5:-5]]
self._show_board(b)
time.sleep(0.1)
class Node:
x = 0
y = 0
rot = 0
final = False
edges = []
def __eq__(self, other):
"""Overrides the default implementation"""
if isinstance(other, Node):
return (self.x == other.x) and (self.y == other.y) and (self.rot == other.rot)
return False
def is_neighbour(self, other):
if (abs(self.x - other.x) + abs(self.y - other.y) + abs(self.rot - other.rot) == 1) and (other.y >= self.y):
return True
return False
def __hash__(self):
return hash((self.x, self.y, self.rot))
def get_possible_moves(model, tile):
start_node = Node()
start_node.x = model._tile.x
start_node.y = model.get_height() - len(tile.shape)-1
frontier = [start_node]
visited = {start_node: True}
final_nodes = []
while len(frontier) > 0:
n = frontier.pop()
for dx, dy, rot in [(-1, 0, 0), (1, 0, 0), (0, 1, 0), (0, 0, 1)][::-1]:
nn = Node()
nn.x = n.x + dx
nn.y = n.y + dy
nn.rot = (n.rot + rot) % 4
if nn not in visited:
visited[nn] = True
t = Tile(tile.id)
t.x = nn.x
t.y = nn.y
for r in range(nn.rot):
t.rotate()
if model._is_valid(model._field, t):
frontier.append(nn)
# check if node is final
for i in range(len(t.shape)):
for j in range(len(t.shape[0])):
if (t.shape[i][j] > 0) and (model._field[nn.y + i + 1][nn.x + j] > 0):
nn.final = True
final_nodes.append(nn)
break
if (nn.final):
break
print(len(visited))
print(len(final_nodes))
return final_nodes
m = Model()
m._field = [
[1, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 1],
[1, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 1],
[1, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 1],
[1, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 1],
[1, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 1],
[1, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 1],
[1, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 1],
[1, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 1],
[1, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 1],
[1, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 1],
[1, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 1],
[1, 1, 1, 1, 1, 0, 3, 0, 0, 0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 1],
[1, 1, 1, 1, 1, 3, 3, 0, 0, 0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 1],
[1, 1, 1, 1, 1, 3, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 1],
[1, 1, 1, 1, 1, 3, 3, 3, 3, 3, 0, 0, 0, 0, 0, 1, 1, 1, 1, 1],
[1, 1, 1, 1, 1, 3, 3, 0, 0, 0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 1],
[1, 1, 1, 1, 1, 3, 3, 0, 3, 0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 1],
[1, 1, 1, 1, 1, 3, 0, 0, 0, 0, 0, 0, 0, 0, 3, 1, 1, 1, 1, 1],
[1, 1, 1, 1, 1, 3, 3, 3, 3, 3, 0, 0, 0, 0, 3, 1, 1, 1, 1, 1],
[1, 1, 1, 1, 1, 3, 3, 3, 3, 3, 0, 0, 0, 3, 3, 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, 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, 1, 1, 1, 1, 1],
]
m._tile = Tile(3)
import threading
import keyboard
# define a thread which takes input
class InputThread(threading.Thread):
def run(self):
#self.daemon = True
self.last_user_input = None
while True:
try:
if keyboard.is_pressed('left'):#if key 'q' is pressed
self.last_user_input = 'a'
m.move_left()
elif keyboard.is_pressed('right'):#if key 'q' is pressed
self.last_user_input = 'a'
m.move_right()
elif keyboard.is_pressed('z'):
print('z')
self.last_user_input = 'z'
m.rotate()
elif keyboard.is_pressed('down'):
m.move_down()
elif keyboard.is_pressed('q'):
break
else:
pass
# do something based on the user input here
# alternatively, let main do something with
# self.last_user_input
except:
pass
time.sleep(0.1)
# main
#it = InputThread()
#it.start()
fn = get_possible_moves(m, m._tile)
time.sleep(2)
v = AsciimaticView(m)
m.set_view(v)
v.show()
time.sleep(2)
for n in fn:
m._tile = Tile(m._tile.id)
m._tile.x = n.x
m._tile.y = n.y
for r in range(n.rot):
m._tile.rotate()
v.show()
time.sleep(1)
time.sleep(500)
Related
The current value is the maximum value in the most recent period
l = [5,1,1,1,5,3,6], and the expected returned data is [0, 0, 0, 0, 3, 0, 6]. Compare from right to left, if it is greater than, it will count cumulatively; if it is less than or equal to, it will interrupt the accumulation and start the next counting. How to implement (numpy,pandas)? pandas: def TOPRANGE(S): rt = np.zeros(len(S)) for i in range(1,len(S)): rt[i] = np.argmin(np.flipud(S[:i]<S[i])) return rt.astype('int') l = [5,1,1,1,5,3,6] s = np.array(l) TOPRANGE(s) output: [0, 0, 0, 0, 3, 0, 0] expected returned data is [0, 0, 0, 0, 3, 0, 6],Don't know how to solve it????
l = [5, 1, 1, 1, 5, 3, 6] # l = [5, 1, 3, 6] l.reverse(); out = [] for i in range(0,len(l)): acc = 0 for j in range(i+1, len(l)): if l[i] > l[j]: acc += 1 else: break # interrupt count out.append(acc) out.reverse(); out Gives required output: [0, 0, 0, 0, 3, 0, 6]
If I understand well the logic, this looks like an expanding comparison: pure python l = [5,1,3,6] out = [sum(l[i]>x for x in l[:i]) for i in range(len(l))] pandas l = [5,1,3,6] s = pd.Series(l) out = s.expanding().apply(lambda x: sum(x.iloc[:-1].le(x.iloc[-1]))).astype(int) out.tolist() numpy: l = [5,1,3,6] a = np.array(l) out = np.tril(a[:,None]>a).sum(1) out.tolist() output: [0, 0, 1, 3]
Identify vectors being a multiple of another in rectangular matrix
Given a nxm matrix (n > m) of integers, I'd like to identify rows that are a multiple of a single other row, so not a linear combination of multiple other rows. I could scale all rows to their length and find unique rows, but that is prone to numerical issues on floating points and would also not detect vectors being opposite (pointing in the other directon) of each other. Any ideas? Example A = array([[-1, -1, 0, 0], [-1, -1, 0, 1], [-1, 0, -1, 0], [-1, 0, 0, 0], [-1, 0, 0, 1], [-1, 0, 1, 1], [-1, 1, -1, 0], [-1, 1, 0, 0], [-1, 1, 1, 0], [ 0, -1, 0, 0], [ 0, -1, 0, 1], [ 0, -1, 1, 0], [ 0, -1, 1, 1], [ 0, 0, -1, 0], [ 0, 0, 0, 1], [ 0, 0, 1, 0], [ 0, 1, -1, 0], [ 0, 1, 0, 0], [ 0, 1, 0, 1], [ 0, 1, 1, 0], [ 0, 1, 1, 1], [ 1, -1, 0, 0], [ 1, -1, 1, 0], [ 1, 0, 0, 0], [ 1, 0, 0, 1], [ 1, 0, 1, 0], [ 1, 0, 1, 1], [ 1, 1, 0, 0], [ 1, 1, 0, 1], [ 1, 1, 1, 0]]) For example Rows 0 and -3 just point in the opposite direction (multiply one by -1 to make them equal).
You can normalize each row dividing it by its GCD: import numpy as np def normalize(a): return a // np.gcd.reduce(a, axis=1, keepdims=True) And you can define a distance that considers opposite vectors as equal: def distance(a, b): equal = np.all(a == b) or np.all(a == -b) return 0 if equal else 1 Then you can use standard clustering methods: from scipy.spatial.distance import pdist from scipy.cluster.hierarchy import linkage, fcluster def cluster(a): norm_a = normalize(a) distances = pdist(norm_a, metric=distance) return fcluster(linkage(distances), t=0.5) For example: >>> A = np.array([( 1, 2, 3, 4), ... ( 0, 2, 4, 8), ... (-1, -2, -3, -4), ... ( 0, 1, 2, 4), ... (-1, 2, -3, 4), ... ( 2, -4, 6, -8)]) >>> cluster(A) array([2, 3, 2, 3, 1, 1], dtype=int32) Interpretation: cluster 1 is formed by rows 4 and 5, cluster 2 by rows 0 and 2, and cluster 3 by rows 1 and 3.
You can take advantage of the fact that inner product of two normalized linearly dependent vectors gives 1 or -1, so the code could look like this: >>> A_normalized = (A.T/np.linalg.norm(A, axis=-1)).T >>> M = np.absolute(np.einsum('ix,jx->ij', A_normalized, A_normalized)) >>> i, j = np.where(np.isclose(M, 1)) >>> i, j = i[i < j], j[i < j] # Remove repetitions >>> print(i, j) output: [ 0 2 3 6 7 9 11 13] [27 25 23 22 21 17 16 15]
Counting zeros in a rolling - numpy array (including NaNs)
I am trying to find a way of Counting zeros in a rolling using numpy array ? Using pandas I can get it using: df['demand'].apply(lambda x: (x == 0).rolling(7).sum()).fillna(0)) or df['demand'].transform(lambda x: x.rolling(7).apply(lambda x: 7 - np.count _nonzero(x))).fillna(0) In numpy, using the code from Here def rolling_window(a, window_size): shape = (a.shape[0] - window_size + 1, window_size) + a.shape[1:] print(shape) strides = (a.strides[0],) + a.strides return np.lib.stride_tricks.as_strided(a, shape=shape, strides=strides) arr = np.asarray([10, 20, 30, 5, 6, 0, 0, 0]) np.count_nonzero(rolling_window(arr==0, 7), axis=1) Output: array([2, 3]) However, I need the first 6 NaNs as well, and fill it with zeros: Expected output: array([0, 0, 0, 0, 0, 0, 2, 3])
Think an efficient one would be with 1D convolution - def sum_occurences_windowed(arr, W): K = np.ones(W, dtype=int) out = np.convolve(arr==0,K)[:len(arr)] out[:W-1] = 0 return out Sample run - In [42]: arr Out[42]: array([10, 20, 30, 5, 6, 0, 0, 0]) In [43]: sum_occurences_windowed(arr,W=7) Out[43]: array([0, 0, 0, 0, 0, 0, 2, 3]) Timings on varying length arrays and window of 7 Including count_rolling from #Quang Hoang's post. Using benchit package (few benchmarking tools packaged together; disclaimer: I am its author) to benchmark proposed solutions. import benchit funcs = [sum_occurences_windowed, count_rolling] in_ = {n:(np.random.randint(0,5,(n)),7) for n in [10,20,50,100,200,500,1000,2000,5000]} t = benchit.timings(funcs, in_, multivar=True, input_name='Length') t.plot(logx=True, save='timings.png') Extending to generic n-dim arrays from scipy.ndimage.filters import convolve1d def sum_occurences_windowed_ndim(arr, W, axis=-1): K = np.ones(W, dtype=int) out = convolve1d((arr==0).astype(int),K,axis=axis,origin=-(W//2)) out.swapaxes(axis,0)[:W-1] = 0 return out So, on a 2D array, for counting along each row, use axis=1 and for cols, axis=0 and so on. Sample run - In [155]: np.random.seed(0) In [156]: a = np.random.randint(0,3,(3,10)) In [157]: a Out[157]: array([[0, 1, 0, 1, 1, 2, 0, 2, 0, 0], [0, 2, 1, 2, 2, 0, 1, 1, 1, 1], [0, 1, 0, 0, 1, 2, 0, 2, 0, 1]]) In [158]: sum_occurences_windowed_ndim(a, W=7) Out[158]: array([[0, 0, 0, 0, 0, 0, 3, 2, 3, 3], [0, 0, 0, 0, 0, 0, 2, 1, 1, 1], [0, 0, 0, 0, 0, 0, 4, 3, 4, 3]]) # Verify with earlier 1D solution In [159]: np.vstack([sum_occurences_windowed(i,7) for i in a]) Out[159]: array([[0, 0, 0, 0, 0, 0, 3, 2, 3, 3], [0, 0, 0, 0, 0, 0, 2, 1, 1, 1], [0, 0, 0, 0, 0, 0, 4, 3, 4, 3]]) Let's test out our original 1D input array - In [187]: arr Out[187]: array([10, 20, 30, 5, 6, 0, 0, 0]) In [188]: sum_occurences_windowed_ndim(arr, W=7) Out[188]: array([0, 0, 0, 0, 0, 0, 2, 3])
I would modify the function as follow: def count_rolling(a, window_size): shape = (a.shape[0] - window_size + 1, window_size) + a.shape[1:] strides = (a.strides[0],) + a.strides rolling = np.lib.stride_tricks.as_strided(a, shape=shape, strides=strides) out = np.zeros_like(a) out[window_size-1:] = (rolling == 0).sum(1) return out arr = np.asarray([10, 20, 30, 5, 6, 0, 0, 0]) count_rolling(arr,7) Output: array([0, 0, 0, 0, 0, 0, 2, 3])
Is there a way to slice out multiple 2D numpy arrays from one 2D numpy array in one batch operation?
I have a numpy array heatmap of shape (img_height, img_width) and another array bboxes of shape (K, 4), where K is a number of bounding boxes. Each bounding box is defined like so: [x_top_left, y_top_left, width, height]. Here's an example of such array: bboxes = np.array([ [0, 0, 4, 7], [3, 4, 3, 4], [7, 2, 3, 7] ]) heatmap is initally filled with zeros. What I need to do is to put value 1 for each bounding box in it's corresponding place. The resulting heatmap should be: heatmap = np.array([ [1, 1, 1, 1, 0, 0, 0, 0, 0, 0, 0], [1, 1, 1, 1, 0, 0, 0, 0, 0, 0, 0], [1, 1, 1, 1, 0, 0, 0, 1, 1, 1, 0], [1, 1, 1, 1, 0, 0, 0, 1, 1, 1, 0], [1, 1, 1, 1, 1, 1, 0, 1, 1, 1, 0], [1, 1, 1, 1, 1, 1, 0, 1, 1, 1, 0], [1, 1, 1, 1, 1, 1, 0, 1, 1, 1, 0], [0, 0, 0, 1, 1, 1, 0, 1, 1, 1, 0], [0, 0, 0, 0, 0, 0, 0, 1, 1, 1, 0], [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0] ]) Important things to note: axis 0 corresponds to image height axis 1 corresponds to image width I've already solved this using python for loop, like so: for bbox in bboxes: # y_top_left:y_top_left + img_height, x_top_left:x_top_left + img_width heatmap[bbox[1] : bbox[1] + bbox[3], bbox[0] : bbox[0] + bbox[2]] = 1 I would like to avoid using python for loops (if it's possible) and be able to do something like this: heatmap[bboxes[:,1] : bboxes[:,1] + bboxes[:,3], bboxes[:,0]:bboxes[:,0] + bboxes[:,2]] = 1 Is there a way of doing such multiple slicing in numpy? I am aware of numpy integer array indexing, but to generate such indices I am also unable to avoid python for loops.
Transform a matrix made of binomial vectors to ranges for consecutive zeros
I am trying to figure out how to do this transformation symbolically in theano a matrix of undetermined size From: [[0, 0, 0, 1, 0, 0, 0, 0, 0, 1, 1, 0, 1, 0, 0, 0, 1], [0, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0, 1, 1, 1, 1, 1, 1], . . ] To: [[1, 2, 3, 0, 1, 2, 3, 4, 5, 0, 0, 1, 0, 1, 2, 3, 0], [1, 2, 3, 4, 5, 6, 7, 8, 0, 1, 2, 0, 0, 0, 0, 0, 0], . . ] So for every consecutive 0 I want an increasing range and whenever I stumble on a 1 the range resets.
Here's one way to do it, using inefficient scans: import theano import theano.tensor as tt def inner_step(x_t_t, y_t_tm1): return tt.switch(x_t_t, 0, y_t_tm1 + 1) def outer_step(x_t): return theano.scan(inner_step, sequences=[x_t], outputs_info=[0])[0] def compile(): x = tt.bmatrix() y = theano.scan(outer_step, sequences=[x])[0] return theano.function([x], y) def main(): f = compile() data = [[0, 0, 0, 1, 0, 0, 0, 0, 0, 1, 1, 0, 1, 0, 0, 0, 1], [0, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0, 1, 1, 1, 1, 1, 1]] print f(data) main() When run, this prints: [[1 2 3 0 1 2 3 4 5 0 0 1 0 1 2 3 0] [1 2 3 4 5 6 7 8 0 1 2 0 0 0 0 0 0]]