def run_experiment_areas_size_diff(self,
number_of_iterations,
number_of_partitions,
s,
grid_base_size=5,
remove_off=False):
smallest_areas_diff = float("inf")
grid_to_draw = None
for i in range(number_of_iterations):
print('ITERATION', i + 1)
grid = Grid(show_progress=False)
grid.load_image(self.grid_name, remove_off=remove_off)
grid.reduce_areas()
grid.reduce_by_lam(number_of_partitions, draw_steps=False)
grid.create_partitions()
grid.fully_restore_with_partitions_improvement(p=0.3)
grid.remove_noises()
max_area, min_area = grid.get_highest_and_smallest_partition()
diff = max_area - min_area
if diff < smallest_areas_diff:
cut_size = grid._get_cut_size()
print('CUT_SIZE:', cut_size)
print('DIFF:', diff)
smallest_areas_diff = diff
grid_to_draw = grid
grid_to_draw.draw_initial_grid(p=1, s=s, base_size=grid_base_size)
grid_to_draw.draw_partitioned_grid(p=1, s=s, base_size=grid_base_size)
grid_to_draw.print_areas_stats()
def run_experiment_areas_size_std(self,
number_of_iterations,
number_of_partitions,
s,
grid_base_size=5,
remove_off=False):
min_std = float("inf")
grid_to_draw = None
for i in range(number_of_iterations):
print('ITERATION', i + 1)
grid = Grid(show_progress=False)
grid.load_image(self.grid_name, remove_off=remove_off)
grid.reduce_areas()
grid.reduce_by_lam(number_of_partitions, draw_steps=False)
grid.create_partitions()
grid.fully_restore_with_partitions_improvement(p=0.3)
grid.remove_noises()
std = numpy.std(numpy.array(list(grid.partitions_stats.values())))
if std < min_std:
cut_size = grid._get_cut_size()
print('CUT_SIZE:', cut_size)
print('STD:', std)
min_std = std
grid_to_draw = grid
grid_to_draw.draw_initial_grid(p=1, s=s, base_size=grid_base_size)
grid_to_draw.draw_partitioned_grid(p=1, s=s, base_size=grid_base_size)
grid_to_draw.print_areas_stats()
def partition_for_computations(self,
number_of_cores,
number_of_nodes,
s,
p=1,
show_progress=False,
grid_base_size=5,
remove_off=False):
grid1 = Grid(show_progress=show_progress)
grid1.load_image(self.grid_name, remove_off=remove_off)
# grid1.draw_initial_grid(p, s)
grid1.reduce_areas()
grid1.reduce_by_lam(number_of_cores * number_of_nodes)
grid1.create_partitions()
grid1.fully_restore_with_partitions_improvement(0.5)
grid1.remove_noises()
# grid1.draw_partitioned_grid(p, s, base_size=grid_base_size)
compact_G = grid1.create_compact_graph()
nodes_partitioner = NodesPartitioner(show_progress=show_progress)
nodes_partitioner.load_partitioned_graph_object(
compact_G=compact_G,
number_of_partitions=number_of_nodes * number_of_cores)
partitions = nodes_partitioner.get_equal_partitioning_by_lam(
number_of_nodes)
print(partitions)
grid1.load_new_partitions(partitions)
grid1.fully_restore()
# grid1.print_areas_stats()
grid1.draw_partitioned_grid(p, s, base_size=grid_base_size)
grid1.print_areas_stats()
def normal_partitioning(self,
number_of_partitions,
s,
p=1,
grid_base_size=5,
show_progress=False,
remove_off=False):
grid = Grid(show_progress=show_progress)
grid.load_image(self.grid_name, remove_off=remove_off)
grid.draw_initial_grid(p, s, optimized=False)
start = time.time()
grid.reduce_areas()
grid.reduce_by_lam(number_of_partitions, draw_steps=False)
grid.create_partitions()
grid.print_areas_stats()
grid._draw_one_step_of_restoration(p, s)
grid.fully_restore_with_partitions_improvement(0.2)
grid.print_execution_times()
print(time.time() - start)
grid.remove_noises()
grid.print_areas_stats()
grid.draw_partitioned_grid(p, s, base_size=grid_base_size)