def where(condition, x=pu.dummy, y=pu.dummy):
a = condition
s = arrayfire.where(a.d_array)
# numpy uses int64 while arrayfire uses uint32
s = ndarray(pu.af_shape(s), dtype=numpy.uint32,
af_array=s).astype(numpy.int64)
# Looks like where goes through the JIT??
s.eval()
if (x is pu.dummy and y is pu.dummy):
idx = []
mult = 1
for i in a.shape[::-1]:
mult = i
idx = [s % mult] + idx
s //= mult
idx = tuple(idx)
return idx
elif (x is not pu.dummy and y is not pu.dummy):
if (x.dtype != y.dtype):
raise TypeError('x and y must have same dtype')
if (x.shape != y.shape):
raise ValueError('x and y must have same shape')
ret = array(y)
if (len(ret.shape) > 1):
ret = ret.flatten()
ret[s] = x.flatten()[s]
ret = ret.reshape(x.shape)
else:
ret[s] = x[s]
return ret
else:
raise ValueError('either both or neither of x and y should be given')
def where(condition, x=pu.dummy, y=pu.dummy):
a = condition
s = arrayfire.where(a.d_array)
# numpy uses int64 while arrayfire uses uint32
s = ndarray(pu.af_shape(s), dtype=numpy.uint32, af_array=s).astype(numpy.int64)
# Looks like where goes through the JIT??
s.eval()
if(x is pu.dummy and y is pu.dummy):
idx = []
mult = 1
for i in a.shape[::-1]:
mult = i
idx = [s % mult] + idx
s //= mult
idx = tuple(idx)
return idx
elif(x is not pu.dummy and y is not pu.dummy):
if(x.dtype != y.dtype):
raise TypeError('x and y must have same dtype')
if(x.shape != y.shape):
raise ValueError('x and y must have same shape')
ret = array(y)
if(len(ret.shape) > 1):
ret = ret.flatten()
ret[s] = x.flatten()[s]
ret = ret.reshape(x.shape)
else:
ret[s] = x[s]
return ret;
else:
raise ValueError('either both or neither of x and y should be given')
def setup_mnist(frac, expand_labels):
root_path = os.path.dirname(os.path.abspath(__file__))
file_path = root_path + '/../../assets/examples/data/mnist/'
idims, idata = read_idx(file_path + 'images-subset')
ldims, ldata = read_idx(file_path + 'labels-subset')
idims.reverse()
numdims = len(idims)
images = af.Array(idata, tuple(idims))
R = af.randu(10000, 1)
cond = R < min(frac, 0.8)
train_indices = af.where(cond)
test_indices = af.where(~cond)
train_images = af.lookup(images, train_indices, 2) / 255
test_images = af.lookup(images, test_indices, 2) / 255
num_classes = 10
num_train = train_images.dims()[2]
num_test = test_images.dims()[2]
if expand_labels:
train_labels = af.constant(0, num_classes, num_train)
test_labels = af.constant(0, num_classes, num_test)
h_train_idx = train_indices.to_list()
h_test_idx = test_indices.to_list()
for i in range(num_train):
train_labels[ldata[h_train_idx[i]], i] = 1
for i in range(num_test):
test_labels[ldata[h_test_idx[i]], i] = 1
else:
labels = af.Array(ldata, tuple(ldims))
train_labels = labels[train_indices]
test_labels = labels[test_indices]
return (num_classes, num_train, num_test, train_images, test_images,
train_labels, test_labels)
def nonzero(self):
s = arrayfire.where(self.d_array)
s = ndarray(pu.af_shape(s), dtype=numpy.uint32,
af_array=s).astype(numpy.int64)
# TODO: Unexplained eval
s.eval()
idx = []
mult = 1
for i in self.shape[::-1]:
mult = i
idx = [s % mult] + idx
s //= mult
idx = tuple(idx)
return idx
def nonzero(self):
s = arrayfire.where(self.d_array)
s = ndarray(pu.af_shape(s), dtype=numpy.uint32,
af_array=s).astype(numpy.int64)
# TODO: Unexplained eval
s.eval()
idx = []
mult = 1
for i in self.shape[::-1]:
mult = i
idx = [s % mult] + idx
s //= mult
idx = tuple(idx)
return idx
def histogram_deposition(current_indices_flat, currents_flat, grid_elements):
'''
function: histogram_deposition(current_indices_flat, currents_flat, grid)
inputs: current_indices_flat, currents_flat, grid_elements
current_indices_flat, currents_flat: They denote the indices and the currents
to be deposited on the flattened current vector.
grid_elements: The number of elements present the matrix/vector representing the
currents.
'''
# setting default indices and current for histogram deposition
indices_fix = af.data.range(grid_elements + 1, dtype=af.Dtype.s64)
currents_fix = 0 * af.data.range(grid_elements + 1, dtype=af.Dtype.f64)
# Concatenating the indices and currents in a single vector
combined_indices_flat = af.join(0, indices_fix, current_indices_flat)
combined_currents_flat = af.join(0, currents_fix, currents_flat)
# Sort by key operation
indices, currents = af.sort_by_key(combined_indices_flat,
combined_currents_flat,
dim=0)
# scan by key operation with default binary addition operation which sums up currents
# for the respective indices
Histogram_scan = af.scan_by_key(indices, currents)
# diff1 operation to determine the uniques indices in the current
diff1_op = af.diff1(indices, dim=0)
# Determining the uniques indices for current deposition
indices_unique = af.where(diff1_op > 0)
# Determining the current vector
J_flat = Histogram_scan[indices_unique]
af.eval(J_flat)
return J_flat
def predict_proba2(self, X):
near_locs, near_dists = af.vision.nearest_neighbour(X, self._data, self._dim, \
self._num_nearest, self._match_type)
weights = self._get_neighbor_weights(near_dists)
top_labels = af.moddims(self._labels[near_locs], \
get_dims(near_locs)[0], get_dims(near_locs)[1])
probs = af.constant(0, X.dims()[0], self._num_classes)
for query_idx in range(X.dims()[0]):
query_weights = weights[:, query_idx]
query_top_labels = top_labels[:, query_idx]
for class_idx in range(self._num_classes):
class_weights = query_weights[af.where(
query_top_labels == class_idx)]
probs[query_idx, class_idx] = af.sum(class_weights)
return probs
def nonzero(a: ndarray) -> tp.Tuple:
index_af_array = af.where(a._af_array)
tmp_array = ndarray(index_af_array)
if a.ndim == 1:
return (tmp_array, )
# return tmp_array
elif a.ndim == 2:
return _division_with_remainder(tmp_array, a.shape[0])
elif a.ndim == 3:
tmp_array2, dimension2 = _division_with_remainder(
tmp_array, a.shape[1])
dimension0, dimension1 \
= _division_with_remainder(tmp_array2, a.shape[0])
return dimension0, dimension1, dimension2
else:
raise ValueError("function nonzero is not implemented for arrays with "
"more than three axes")
def indices_and_currents_TSC_2D( charge_electron, positions_x, positions_y, velocity_x, velocity_y,\
x_grid, y_grid, ghost_cells, length_domain_x, length_domain_y, dt ):
positions_x_new = positions_x + velocity_x * dt
positions_y_new = positions_y + velocity_y * dt
base_indices_x = af.data.constant(0, positions_x.elements(), dtype=af.Dtype.u32)
base_indices_y = af.data.constant(0, positions_x.elements(), dtype=af.Dtype.u32)
dx = af.sum(x_grid[1] - x_grid[0])
dy = af.sum(y_grid[1] - y_grid[0])
# S0
############################################################################################################
x_zone = (((af.abs(positions_x - af.sum(x_grid[0])))/dx).as_type(af.Dtype.u32))
y_zone = (((af.abs(positions_y - af.sum(y_grid[0])))/dy).as_type(af.Dtype.u32))
temp = af.where(af.abs(positions_x-x_grid[x_zone])<af.abs(positions_x-x_grid[x_zone + 1]))
if(temp.elements()>0):
base_indices_x[temp] = x_zone[temp]
temp = af.where(af.abs(positions_x - x_grid[x_zone])>=af.abs(positions_x-x_grid[x_zone + 1]))
if(temp.elements()>0):
base_indices_x[temp] = (x_zone[temp] + 1).as_type(af.Dtype.u32)
temp = af.where(af.abs(positions_y-y_grid[y_zone])<af.abs(positions_y-y_grid[y_zone + 1]))
if(temp.elements()>0):
base_indices_y[temp] = y_zone[temp]
temp = af.where(af.abs(positions_y - y_grid[y_zone])>=af.abs(positions_y-x_grid[y_zone + 1]))
if(temp.elements()>0):
base_indices_y[temp] = (y_zone[temp] + 1).as_type(af.Dtype.u32)
base_indices_minus_two = (base_indices_x - 2).as_type(af.Dtype.u32)
base_indices_minus = (base_indices_x - 1).as_type(af.Dtype.u32)
base_indices_plus = (base_indices_x + 1).as_type(af.Dtype.u32)
base_indices_plus_two = (base_indices_x + 2).as_type(af.Dtype.u32)
index_list_x = af.join( 1,\
af.join(1, base_indices_minus_two, base_indices_minus, base_indices_x),\
af.join(1, base_indices_plus, base_indices_plus_two),\
)
base_indices_minus_two = (base_indices_y - 2).as_type(af.Dtype.u32)
base_indices_minus = (base_indices_y - 1).as_type(af.Dtype.u32)
base_indices_plus = (base_indices_y + 1).as_type(af.Dtype.u32)
base_indices_plus_two = (base_indices_y + 2).as_type(af.Dtype.u32)
index_list_y = af.join( 1,\
af.join(1, base_indices_minus_two, base_indices_minus, base_indices_y),\
af.join(1, base_indices_plus, base_indices_plus_two),\
)
positions_x_5x = af.join( 0,\
af.join(0, positions_x, positions_x, positions_x),\
af.join(0, positions_x, positions_x),\
)
positions_y_5x = af.join( 0,\
af.join(0, positions_y, positions_y, positions_y),\
af.join(0, positions_y, positions_y),\
)
# Determining S0 for positions at t = n * dt
distance_nodes_x = x_grid[af.flat(index_list_x)]
distance_nodes_y = y_grid[af.flat(index_list_y)]
W_x = 0 * distance_nodes_x.copy()
W_y = 0 * distance_nodes_y.copy()
# Determining weights in x direction
temp = af.where(af.abs(distance_nodes_x - positions_x_5x) < (0.5*dx) )
if(temp.elements()>0):
W_x[temp] = 0.75 - (af.abs(distance_nodes_x[temp] - positions_x_5x[temp])/dx)**2
temp = af.where((af.abs(distance_nodes_x - positions_x_5x) >= (0.5*dx) )\
* (af.abs(distance_nodes_x - positions_x_5x) < (1.5 * dx) )\
)
if(temp.elements()>0):
W_x[temp] = 0.5 * (1.5 - (af.abs(distance_nodes_x[temp] - positions_x_5x[temp])/dx))**2
# Determining weights in y direction
temp = af.where(af.abs(distance_nodes_y - positions_y_5x) < (0.5*dy) )
if(temp.elements()>0):
W_y[temp] = 0.75 - (af.abs(distance_nodes_y[temp] - positions_y_5x[temp])/dy)**2
temp = af.where((af.abs(distance_nodes_y - positions_y_5x) >= (0.5*dy) )\
* (af.abs(distance_nodes_y - positions_y_5x) < (1.5 * dy) )\
)
if(temp.elements()>0):
W_y[temp] = 0.5 * (1.5 - (af.abs(distance_nodes_y[temp] - positions_y_5x[temp])/dy))**2
W_x = af.data.moddims(W_x, positions_x.elements(), 5)
W_y = af.data.moddims(W_y, positions_y.elements(), 5)
# print('S0 W_y is ', W_y)
S0_x = af.tile(W_x, 1, 1, 5)
S0_y = af.tile(W_y, 1, 1, 5)
S0_y = af.reorder(S0_y, 0, 2, 1)
#S1
###################################################################################################
positions_x_5x_new = af.join( 0,\
af.join(0, positions_x_new, positions_x_new, positions_x_new),\
af.join(0, positions_x_new, positions_x_new),\
)
positions_y_5x_new = af.join( 0,\
af.join(0, positions_y_new, positions_y_new, positions_y_new),\
af.join(0, positions_y_new, positions_y_new),\
)
# Determining S0 for positions at t = n * dt
W_x = 0 * distance_nodes_x.copy()
W_y = 0 * distance_nodes_y.copy()
# Determining weights in x direction
temp = af.where(af.abs(distance_nodes_x - positions_x_5x_new) < (0.5*dx) )
if(temp.elements()>0):
W_x[temp] = 0.75 - (af.abs(distance_nodes_x[temp] - positions_x_5x_new[temp])/dx)**2
temp = af.where((af.abs(distance_nodes_x - positions_x_5x_new) >= (0.5*dx) )\
* (af.abs(distance_nodes_x - positions_x_5x_new) < (1.5 * dx) )\
)
if(temp.elements()>0):
W_x[temp] = 0.5 * (1.5 - (af.abs(distance_nodes_x[temp] - positions_x_5x_new[temp])/dx))**2
# Determining weights in y direction
temp = af.where(af.abs(distance_nodes_y - positions_y_5x_new) < (0.5*dy) )
if(temp.elements()>0):
W_y[temp] = 0.75 - (af.abs(distance_nodes_y[temp] - positions_y_5x_new[temp])/dy)**2
temp = af.where((af.abs(distance_nodes_y - positions_y_5x_new) >= (0.5*dy) )\
* (af.abs(distance_nodes_y - positions_y_5x_new) < (1.5 * dy) )\
)
if(temp.elements()>0):
W_y[temp] = 0.5 * (1.5 - (af.abs(distance_nodes_y[temp] - positions_y_5x_new[temp])/dy))**2
W_x = af.data.moddims(W_x, positions_x.elements(), 5)
W_y = af.data.moddims(W_y, positions_x.elements(), 5)
# print('S1 W_y is ', W_y)
S1_x = af.tile(W_x, 1, 1, 5)
S1_y = af.tile(W_y, 1, 1, 5)
S1_y = af.reorder(S1_y, 0, 2, 1)
# print('S1_x is ', S1_x)
# print('S0_x is ', S0_x)
# print('S1_y is ', S1_y)
# print('S0_y is ', S0_y)
###############################################################################################
# Determining the final weight matrix in 3D
W_x = (S1_x - S0_x) * (S0_y + (0.5 *(S1_y - S0_y)) )
W_y = (S1_y - S0_y) * (S0_x + (0.5 *(S1_x - S0_x)) )
###############################################################################################
Jx = af.data.constant(0, positions_x.elements(), 5, 5, dtype = af.Dtype.f64)
Jy = af.data.constant(0, positions_x.elements(), 5, 5, dtype = af.Dtype.f64)
Jx[:, 0, :] = -1 * charge_electron * (dx/dt) * W_x[:, 0, :].copy()
Jx[:, 1, :] = Jx[:, 0, :] + -1 * charge_electron * (dx/dt) * W_x[:, 1, :].copy()
Jx[:, 2, :] = Jx[:, 1, :] + -1 * charge_electron * (dx/dt) * W_x[:, 2, :].copy()
Jx[:, 3, :] = Jx[:, 2, :] + -1 * charge_electron * (dx/dt) * W_x[:, 3, :].copy()
Jx[:, 4, :] = Jx[:, 3, :] + -1 * charge_electron * (dx/dt) * W_x[:, 4, :].copy()
Jx = (1/(dx * dy)) * Jx
Jy[:, :, 0] = -1 * charge_electron * (dy/dt) * W_y[:, :, 0].copy()
Jy[:, :, 1] = Jy[:, :, 0] + -1 * charge_electron * (dy/dt) * W_y[:, :, 1].copy()
Jy[:, :, 2] = Jy[:, :, 1] + -1 * charge_electron * (dy/dt) * W_y[:, :, 2].copy()
Jy[:, :, 3] = Jy[:, :, 2] + -1 * charge_electron * (dy/dt) * W_y[:, :, 3].copy()
Jy[:, :, 4] = Jy[:, :, 3] + -1 * charge_electron * (dy/dt) * W_y[:, :, 4].copy()
Jy = (1/(dx * dy)) * Jy
# Determining the x indices for charge deposition
index_list_x_Jx = af.flat(af.tile(index_list_x, 1, 1, 5))
# Determining the y indices for charge deposition
y_current_zone = af.tile(index_list_y, 1, 1, 5)
index_list_y_Jx = af.flat(af.reorder(y_current_zone, 0, 2, 1))
currents_Jx = af.flat(Jx)
# Determining the x indices for charge deposition
index_list_x_Jy = af.flat(af.tile(index_list_x, 1, 1, 5))
# Determining the y indices for charge deposition
y_current_zone = af.tile(index_list_y, 1, 1, 5)
index_list_y_Jy = af.flat(af.reorder(y_current_zone, 0, 2, 1))
currents_Jy = af.flat(Jy)
af.eval(index_list_x_Jx, index_list_y_Jx)
af.eval(index_list_x_Jy, index_list_y_Jy)
af.eval(currents_Jx, currents_Jy)
return index_list_x_Jx, index_list_y_Jx, currents_Jx,\
index_list_x_Jy, index_list_y_Jy,\
currents_Jy
def flatnonzero(a: ndarray) -> ndarray:
return ndarray(af.where(a._af_array))
def transform(self, X):
"""Impute all missing values in X.
Parameters
----------
X : {array-like, sparse matrix}, shape (n_samples, n_features)
The input data to complete.
"""
check_is_fitted(self)
X = self._validate_input(X, in_fit=False)
#X = af.Array.to_ndarray(X)
X_indicator = super()._transform_indicator(X)
statistics = self.statistics_
if X.shape[1] != statistics.shape[0]:
raise ValueError(
f"X has {X.shape[1]} features per sample, expected {self.statistics_.shape[0]}"
)
# Delete the invalid columns if strategy is not constant
if self.strategy == "constant":
valid_statistics = statistics
else:
# same as af.isnan but also works for object dtypes
# invalid_mask = _get_mask(statistics, np.nan) # BUG: af runtime error
invalid_mask = af.isnan(statistics) # FIXME
valid_mask = invalid_mask.logical_not()
valid_statistics = statistics[valid_mask]
valid_statistics_indexes = np.flatnonzero(valid_mask)
if af.any_true(invalid_mask):
missing = af.arange(X.shape[1])[invalid_mask]
if self.verbose:
warnings.warn(
f"Deleting features without observed values: {missing}"
)
X = X[:, valid_statistics_indexes]
# Do actual imputation
if sp.issparse(X):
if self.missing_values == 0:
raise ValueError(
"Imputation not possible when missing_values == 0 and input is sparse."
"Provide a dense array instead.")
else:
mask = _get_mask(X.data, self.missing_values)
indexes = af.repeat(af.arange(len(X.indptr) - 1, dtype=af.int),
af.diff(X.indptr))[mask]
X.data[mask] = valid_statistics[indexes].astype(X.dtype,
copy=False)
else:
# mask = _get_mask(X, self.missing_values) # BUG
mask = af.isnan(X) # FIXME
# n_missing = af.sum(mask, axis=0) # BUG af
n_missing = af.sum(mask, dim=0)
coordinates = af.where(mask.T)[::-1] # BUG
valid_statistics = valid_statistics.to_ndarray().ravel()
n_missing = n_missing.to_ndarray().ravel()
values = np.repeat(valid_statistics, n_missing) # BUG
values = af.interop.from_ndarray(values)
odims = X.dims()
X = af.flat(X)
X[coordinates] = values
X = af.moddims(X, *odims)
return super()._concatenate_indicator(X, X_indicator)
def indices_and_currents_TSC_2D( charge_electron, positions_x, positions_y, velocity_x, velocity_y,\
x_grid, y_grid, ghost_cells, length_domain_x, length_domain_y, dt ):
"""
function indices_and_currents_TSC_2D( charge_electron, positions_x,\
positions_y, velocity_x, velocity_y,\
x_grid, y_grid, ghost_cells,\
length_domain_x, length_domain_y, dt\
)
return the x and y indices for Jx and Jy and respective currents associated with these indices
"""
positions_x_new = positions_x + velocity_x * dt
positions_y_new = positions_y + velocity_y * dt
base_indices_x = af.data.constant(0,
positions_x.elements(),
dtype=af.Dtype.u32)
base_indices_y = af.data.constant(0,
positions_x.elements(),
dtype=af.Dtype.u32)
dx = af.sum(x_grid[1] - x_grid[0])
dy = af.sum(y_grid[1] - y_grid[0])
# Computing S0_x and S0_y
###########################################################################################
# Determining the grid cells containing the respective particles
x_zone = (((af.abs(positions_x - af.sum(x_grid[0]))) / dx).as_type(
af.Dtype.u32))
y_zone = (((af.abs(positions_y - af.sum(y_grid[0]))) / dy).as_type(
af.Dtype.u32))
# Determing the indices of the closest grid node in x direction
temp = af.where(af.abs(positions_x-x_grid[x_zone]) < \
af.abs(positions_x-x_grid[x_zone + 1])\
)
if (temp.elements() > 0):
base_indices_x[temp] = x_zone[temp]
temp = af.where(af.abs(positions_x - x_grid[x_zone]) >= \
af.abs(positions_x-x_grid[x_zone + 1])\
)
if (temp.elements() > 0):
base_indices_x[temp] = (x_zone[temp] + 1).as_type(af.Dtype.u32)
# Determing the indices of the closest grid node in y direction
temp = af.where(af.abs(positions_y-y_grid[y_zone]) < \
af.abs(positions_y-y_grid[y_zone + 1])\
)
if (temp.elements() > 0):
base_indices_y[temp] = y_zone[temp]
temp = af.where(
af.abs(positions_y - y_grid[y_zone]) >= af.abs(positions_y -
x_grid[y_zone + 1]))
if (temp.elements() > 0):
base_indices_y[temp] = (y_zone[temp] + 1).as_type(af.Dtype.u32)
# Concatenating the index list for near by grid nodes in x direction
# TSC affect 5 nearest grid nodes around in 1 Dimensions
base_indices_minus_two = (base_indices_x - 2).as_type(af.Dtype.u32)
base_indices_minus = (base_indices_x - 1).as_type(af.Dtype.u32)
base_indices_plus = (base_indices_x + 1).as_type(af.Dtype.u32)
base_indices_plus_two = (base_indices_x + 2).as_type(af.Dtype.u32)
index_list_x = af.join( 1,\
af.join(1, base_indices_minus_two, base_indices_minus, base_indices_x),\
af.join(1, base_indices_plus, base_indices_plus_two),\
)
# Concatenating the index list for near by grid nodes in y direction
# TSC affect 5 nearest grid nodes around in 1 Dimensions
base_indices_minus_two = (base_indices_y - 2).as_type(af.Dtype.u32)
base_indices_minus = (base_indices_y - 1).as_type(af.Dtype.u32)
base_indices_plus = (base_indices_y + 1).as_type(af.Dtype.u32)
base_indices_plus_two = (base_indices_y + 2).as_type(af.Dtype.u32)
index_list_y = af.join( 1,\
af.join(1, base_indices_minus_two, base_indices_minus, base_indices_y),\
af.join(1, base_indices_plus, base_indices_plus_two),\
)
# Concatenating the positions_x for determining weights for near by grid nodes in y direction
# TSC affect 5 nearest grid nodes around in 1 Dimensions
positions_x_5x = af.join( 0,\
af.join(0, positions_x, positions_x, positions_x),\
af.join(0, positions_x, positions_x),\
)
positions_y_5x = af.join( 0,\
af.join(0, positions_y, positions_y, positions_y),\
af.join(0, positions_y, positions_y),\
)
# Determining S0 for positions at t = n * dt
distance_nodes_x = x_grid[af.flat(index_list_x)]
distance_nodes_y = y_grid[af.flat(index_list_y)]
W_x = 0 * distance_nodes_x.copy()
W_y = 0 * distance_nodes_y.copy()
# Determining weights in x direction
temp = af.where(af.abs(distance_nodes_x - positions_x_5x) < (0.5 * dx))
if (temp.elements() > 0):
W_x[temp] = 0.75 - (
af.abs(distance_nodes_x[temp] - positions_x_5x[temp]) / dx)**2
temp = af.where((af.abs(distance_nodes_x - positions_x_5x) >= (0.5*dx) )\
* (af.abs(distance_nodes_x - positions_x_5x) < (1.5 * dx) )\
)
if (temp.elements() > 0):
W_x[temp] = 0.5 * (
1.5 -
(af.abs(distance_nodes_x[temp] - positions_x_5x[temp]) / dx))**2
# Determining weights in y direction
temp = af.where(af.abs(distance_nodes_y - positions_y_5x) < (0.5 * dy))
if (temp.elements() > 0):
W_y[temp] = 0.75 - (
af.abs(distance_nodes_y[temp] - positions_y_5x[temp]) / dy)**2
temp = af.where((af.abs(distance_nodes_y - positions_y_5x) >= (0.5*dy) )\
* (af.abs(distance_nodes_y - positions_y_5x) < (1.5 * dy) )\
)
if (temp.elements() > 0):
W_y[temp] = 0.5 * (
1.5 -
(af.abs(distance_nodes_y[temp] - positions_y_5x[temp]) / dy))**2
# Restructering W_x and W_y for visualization and ease of understanding
W_x = af.data.moddims(W_x, positions_x.elements(), 5)
W_y = af.data.moddims(W_y, positions_y.elements(), 5)
# Tiling the S0_x and S0_y for the 25 indices around the particle
S0_x = af.tile(W_x, 1, 1, 5)
S0_y = af.tile(W_y, 1, 1, 5)
S0_y = af.reorder(S0_y, 0, 2, 1)
#Computing S1_x and S1_y
###########################################################################################
positions_x_5x_new = af.join( 0,\
af.join(0, positions_x_new, positions_x_new, positions_x_new),\
af.join(0, positions_x_new, positions_x_new),\
)
positions_y_5x_new = af.join( 0,\
af.join(0, positions_y_new, positions_y_new, positions_y_new),\
af.join(0, positions_y_new, positions_y_new),\
)
# Determining S0 for positions at t = n * dt
W_x = 0 * distance_nodes_x.copy()
W_y = 0 * distance_nodes_y.copy()
# Determining weights in x direction
temp = af.where(af.abs(distance_nodes_x - positions_x_5x_new) < (0.5 * dx))
if (temp.elements() > 0):
W_x[temp] = 0.75 - (
af.abs(distance_nodes_x[temp] - positions_x_5x_new[temp]) / dx)**2
temp = af.where((af.abs(distance_nodes_x - positions_x_5x_new) >= (0.5*dx) )\
* (af.abs(distance_nodes_x - positions_x_5x_new) < (1.5 * dx) )\
)
if (temp.elements() > 0):
W_x[temp] = 0.5 * (1.5 - (af.abs(distance_nodes_x[temp] \
- positions_x_5x_new[temp])/dx\
)\
)**2
# Determining weights in y direction
temp = af.where(af.abs(distance_nodes_y - positions_y_5x_new) < (0.5 * dy))
if (temp.elements() > 0):
W_y[temp] = 0.75 - (af.abs(distance_nodes_y[temp] \
- positions_y_5x_new[temp]\
)/dy\
)**2
temp = af.where((af.abs(distance_nodes_y - positions_y_5x_new) >= (0.5*dy) )\
* (af.abs(distance_nodes_y - positions_y_5x_new) < (1.5 * dy) )\
)
if (temp.elements() > 0):
W_y[temp] = 0.5 * (1.5 - (af.abs(distance_nodes_y[temp] \
- positions_y_5x_new[temp])/dy\
)\
)**2
# Restructering W_x and W_y for visualization and ease of understanding
W_x = af.data.moddims(W_x, positions_x.elements(), 5)
W_y = af.data.moddims(W_y, positions_x.elements(), 5)
# Tiling the S0_x and S0_y for the 25 indices around the particle
S1_x = af.tile(W_x, 1, 1, 5)
S1_y = af.tile(W_y, 1, 1, 5)
S1_y = af.reorder(S1_y, 0, 2, 1)
###########################################################################################
# Determining the final weight matrix for currents in 3D matrix form factor
W_x = (S1_x - S0_x) * (S0_y + (0.5 * (S1_y - S0_y)))
W_y = (S1_y - S0_y) * (S0_x + (0.5 * (S1_x - S0_x)))
###########################################################################################
# Assigning Jx and Jy according to Esirkepov's scheme
Jx = af.data.constant(0, positions_x.elements(), 5, 5, dtype=af.Dtype.f64)
Jy = af.data.constant(0, positions_x.elements(), 5, 5, dtype=af.Dtype.f64)
Jx[:, 0, :] = -1 * charge_electron * (dx / dt) * W_x[:, 0, :].copy()
Jx[:,
1, :] = Jx[:, 0, :] + -1 * charge_electron * (dx /
dt) * W_x[:, 1, :].copy()
Jx[:,
2, :] = Jx[:, 1, :] + -1 * charge_electron * (dx /
dt) * W_x[:, 2, :].copy()
Jx[:,
3, :] = Jx[:, 2, :] + -1 * charge_electron * (dx /
dt) * W_x[:, 3, :].copy()
Jx[:,
4, :] = Jx[:, 3, :] + -1 * charge_electron * (dx /
dt) * W_x[:, 4, :].copy()
# Computing current density using currents
Jx = (1 / (dx * dy)) * Jx
Jy[:, :, 0] = -1 * charge_electron * (dy / dt) * W_y[:, :, 0].copy()
Jy[:, :,
1] = Jy[:, :,
0] + -1 * charge_electron * (dy / dt) * W_y[:, :, 1].copy()
Jy[:, :,
2] = Jy[:, :,
1] + -1 * charge_electron * (dy / dt) * W_y[:, :, 2].copy()
Jy[:, :,
3] = Jy[:, :,
2] + -1 * charge_electron * (dy / dt) * W_y[:, :, 3].copy()
Jy[:, :,
4] = Jy[:, :,
3] + -1 * charge_electron * (dy / dt) * W_y[:, :, 4].copy()
# Computing current density using currents
Jy = (1 / (dx * dy)) * Jy
# Preparing the final index and current vectors
###########################################################################################
# Determining the x indices for charge deposition
index_list_x_Jx = af.flat(af.tile(index_list_x, 1, 1, 5))
# Determining the y indices for charge deposition
y_current_zone = af.tile(index_list_y, 1, 1, 5)
index_list_y_Jx = af.flat(af.reorder(y_current_zone, 0, 2, 1))
currents_Jx = af.flat(Jx)
# Determining the x indices for charge deposition
index_list_x_Jy = af.flat(af.tile(index_list_x, 1, 1, 5))
# Determining the y indices for charge deposition
y_current_zone = af.tile(index_list_y, 1, 1, 5)
index_list_y_Jy = af.flat(af.reorder(y_current_zone, 0, 2, 1))
# Flattenning the Currents array
currents_Jy = af.flat(Jy)
af.eval(index_list_x_Jx, index_list_y_Jx)
af.eval(index_list_x_Jy, index_list_y_Jy)
af.eval(currents_Jx, currents_Jy)
return index_list_x_Jx, index_list_y_Jx, currents_Jx,\
index_list_x_Jy, index_list_y_Jy, currents_Jy
def harris_demo(console):
root_path = os.path.dirname(os.path.abspath(__file__))
file_path = root_path
if console:
file_path += "/../../assets/examples/images/square.png"
else:
file_path += "/../../assets/examples/images/man.jpg"
img_color = af.load_image(file_path, True)
img = af.color_space(img_color, af.CSPACE.GRAY, af.CSPACE.RGB)
img_color /= 255.0
ix, iy = af.gradient(img)
ixx = ix * ix
ixy = ix * iy
iyy = iy * iy
# Compute a Gaussian kernel with standard deviation of 1.0 and length of 5 pixels
# These values can be changed to use a smaller or larger window
gauss_filt = af.gaussian_kernel(5, 5, 1.0, 1.0)
# Filter second order derivatives
ixx = af.convolve(ixx, gauss_filt)
ixy = af.convolve(ixy, gauss_filt)
iyy = af.convolve(iyy, gauss_filt)
# Calculate trace
itr = ixx + iyy
# Calculate determinant
idet = ixx * iyy - ixy * ixy
# Calculate Harris response
response = idet - 0.04 * (itr * itr)
# Get maximum response for each 3x3 neighborhood
mask = af.constant(1, 3, 3)
max_resp = af.dilate(response, mask)
# Discard responses that are not greater than threshold
corners = response > 1e5
corners = corners * response
# Discard responses that are not equal to maximum neighborhood response,
# scale them to original value
corners = (corners == max_resp) * corners
# Copy device array to python list on host
corners_list = corners.to_list()
draw_len = 3
good_corners = 0
for x in range(img_color.dims()[1]):
for y in range(img_color.dims()[0]):
if corners_list[x][y] > 1e5:
img_color = draw_corners(img_color, x, y, draw_len)
good_corners += 1
print("Corners found: {}".format(good_corners))
if not console:
# Previews color image with green crosshairs
wnd = af.Window(512, 512, "Harris Feature Detector")
while not wnd.close():
wnd.image(img_color)
else:
idx = af.where(corners)
corners_x = idx / float(corners.dims()[0])
corners_y = idx % float(corners.dims()[0])
print(corners_x)
print(corners_y)
def TSC_charge_deposition_2D(charge_electron,\
positions_x, positions_y,\
x_grid, y_grid,\
ghost_cells,\
length_domain_x, length_domain_y\
):
'''
function cloud_charge_deposition(charge, zone_x, frac_x, x_grid, dx)
-----------------------------------------------------------------------
Input variables: charge, zone_x, frac_x, x_grid, dx
charge: This is a scalar denoting the charge of the macro particle in the PIC code.
zone_x: This is an array of size (number of electrons/macro particles) containing the indices
of the left corners of the grid cells containing the respective particles
frac_x: This is an array of size (number of electrons/macro particles) containing the fractional values of
the positions of particles in their respective grid cells. This is used for linear interpolation
x_grid: This is an array denoting the position grid chosen in the PIC simulation
dx: This is the distance between any two consecutive grid nodes of the position grid.
-----------------------------------------------------------------------
returns: rho
rho: This is an array containing the charges deposited at the density grid nodes.
'''
base_indices_x = af.data.constant(0, positions_x.elements(), dtype=af.Dtype.u32)
base_indices_y = af.data.constant(0, positions_x.elements(), dtype=af.Dtype.u32)
dx = af.sum(x_grid[1] - x_grid[0])
dy = af.sum(y_grid[1] - y_grid[0])
x_zone = (((af.abs(positions_x - af.sum(x_grid[0])))/dx).as_type(af.Dtype.u32))
y_zone = (((af.abs(positions_y - af.sum(y_grid[0])))/dy).as_type(af.Dtype.u32))
temp = af.where(af.abs(positions_x-x_grid[x_zone])<af.abs(positions_x-x_grid[x_zone + 1]))
if(temp.elements()>0):
base_indices_x[temp] = x_zone[temp]
temp = af.where(af.abs(positions_x - x_grid[x_zone])>=af.abs(positions_x-x_grid[x_zone + 1]))
if(temp.elements()>0):
base_indices_x[temp] = (x_zone[temp] + 1).as_type(af.Dtype.u32)
temp = af.where(af.abs(positions_y-y_grid[y_zone])<af.abs(positions_y-y_grid[y_zone + 1]))
if(temp.elements()>0):
base_indices_y[temp] = y_zone[temp]
temp = af.where(af.abs(positions_y - y_grid[y_zone])>=af.abs(positions_y-x_grid[y_zone + 1]))
if(temp.elements()>0):
base_indices_y[temp] = (y_zone[temp] + 1).as_type(af.Dtype.u32)
base_indices_minus = (base_indices_x - 1).as_type(af.Dtype.u32)
base_indices_plus = (base_indices_x + 1).as_type(af.Dtype.u32)
index_list_x = af.join( 1, base_indices_minus, base_indices_x, base_indices_plus)
base_indices_minus = (base_indices_y - 1).as_type(af.Dtype.u32)
base_indices_plus = (base_indices_y + 1).as_type(af.Dtype.u32)
index_list_y = af.join( 1, base_indices_minus, base_indices_y, base_indices_plus)
positions_x_3x = af.join( 0,positions_x, positions_x, positions_x)
positions_y_3x = af.join( 0,positions_y, positions_y, positions_y)
distance_nodes_x = x_grid[af.flat(index_list_x)]
distance_nodes_y = y_grid[af.flat(index_list_y)]
W_x = 0 * distance_nodes_x.copy()
W_y = 0 * distance_nodes_y.copy()
# Determining weights in x direction
temp = af.where(af.abs(distance_nodes_x - positions_x_3x) < (0.5*dx) )
if(temp.elements()>0):
W_x[temp] = 0.75 - (af.abs(distance_nodes_x[temp] - positions_x_3x[temp])/dx)**2
temp = af.where((af.abs(distance_nodes_x - positions_x_3x) >= (0.5*dx) )\
* (af.abs(distance_nodes_x - positions_x_3x) < (1.5 * dx) )\
)
if(temp.elements()>0):
W_x[temp] = 0.5 * (1.5 - (af.abs(distance_nodes_x[temp] - positions_x_3x[temp])/dx))**2
# Determining weights in y direction
temp = af.where(af.abs(distance_nodes_y - positions_y_3x) < (0.5*dx) )
if(temp.elements()>0):
W_y[temp] = 0.75 - (af.abs(distance_nodes_y[temp] - positions_y_3x[temp])/dx)**2
temp = af.where((af.abs(distance_nodes_y - positions_y_3x) >= (0.5*dx) )\
* (af.abs(distance_nodes_y - positions_y_3x) < (1.5 * dx) )\
)
if(temp.elements()>0):
W_y[temp] = 0.5 * (1.5 - (af.abs(distance_nodes_y[temp] - positions_y_3x[temp])/dx))**2
W_x = af.data.moddims(W_x, positions_x.elements(), 3)
W_y = af.data.moddims(W_y, positions_x.elements(), 3)
W_x = af.tile(W_x, 1, 1, 3)
W_y = af.tile(W_y, 1, 1, 3)
W_y = af.reorder(W_y, 0, 2, 1)
# Determining the final weight matrix
W = W_x * W_y
W = af.flat(W)
# Determining the x indices for charge deposition
x_charge_zone = af.flat(af.tile(index_list_x, 1, 1, 3))
# Determining the y indices for charge deposition
y_charge_zone = af.tile(index_list_y, 1, 1, 3)
y_charge_zone = af.flat(af.reorder(y_charge_zone, 0, 2, 1))
# Determining the charges to depositied for the respective x and y indices
charges = (charge_electron * W)/(dx * dy)
af.eval(x_charge_zone, y_charge_zone)
af.eval(charges)
return x_charge_zone, y_charge_zone, charges
def indices_and_currents_TSC(charge_electron, positions_x, dt, x_grid, dx,
ghost_cells):
positions_x_new = positions_x + velocity_x * dt
number_of_electrons = positions_x.elements()
base_indices = af.data.constant(0,
positions_x.elements(),
dtype=af.Dtype.u32)
Jx = af.data.constant(0, positions_x.elements(), 5, dtype=af.Dtype.f64)
x_zone = (((af.abs(positions_x - af.sum(x_grid[0]))) / dx).as_type(
af.Dtype.u32))
temp = af.where(
af.abs(positions_x - x_grid[x_zone]) < af.abs(positions_x -
x_grid[x_zone + 1]))
if (temp.elements() > 0):
base_indices[temp] = x_zone[temp]
temp = af.where(
af.abs(positions_x - x_grid[x_zone]) >= af.abs(positions_x -
x_grid[x_zone + 1]))
if (temp.elements() > 0):
base_indices[temp] = (x_zone[temp] + 1).as_type(af.Dtype.u32)
S0 = af.data.constant(0, 5 * positions_x.elements(), dtype=af.Dtype.f64)
S1 = S0.copy()
base_indices_plus = (base_indices + 1).as_type(af.Dtype.u32)
base_indices_plus_two = (base_indices + 2).as_type(af.Dtype.u32)
base_indices_minus = (base_indices - 1).as_type(af.Dtype.u32)
base_indices_minus_two = (base_indices - 2).as_type(af.Dtype.u32)
index_list_1 = af.join( 0, base_indices_minus_two, base_indices_minus,\
)
index_list_2 = af.join( 0, base_indices, base_indices_plus,base_indices_plus_two\
)
index_list = af.join(0, index_list_1, index_list_2)
# Computing S(base_indices - 1, base_indices, base_indices + 1)
positions_x_5x = af.join(0, positions_x, positions_x, positions_x)
positions_x_5x = af.join(0, positions_x_5x, positions_x, positions_x)
positions_x_new_5x = af.join(0, positions_x_new, positions_x_new,
positions_x_new)
positions_x_new_5x = af.join(0, positions_x_new_5x, positions_x_new,
positions_x_new)
# Computing S0
temp = af.where(af.abs(positions_x_5x - x_grid[index_list]) < dx / 2)
if (temp.elements() > 0):
S0[temp] = (0.75) - (
af.abs(positions_x_5x[temp] - x_grid[index_list[temp]]) / dx)**2
temp = af.where((af.abs(positions_x_5x - x_grid[index_list]) >= 0.5 * dx)\
* (af.abs(positions_x_5x - x_grid[index_list]) < 1.5 * dx)\
)
if (temp.elements() > 0):
S0[temp] = 0.5 * (3 / 2 - af.abs(
(positions_x_5x[temp] - x_grid[index_list[temp]]) / dx))**2
# Computing S1
temp = af.where(af.abs(positions_x_new_5x - x_grid[index_list]) < dx / 2)
if (temp.elements() > 0):
S1[temp] = (3 / 4) - (af.abs(positions_x_new_5x[temp] -
x_grid[index_list[temp]]) / dx)**2
temp = af.where((af.abs(positions_x_new_5x - x_grid[index_list]) >= 0.5 * dx)\
* (af.abs(positions_x_new_5x - x_grid[index_list]) < 1.5 * dx)\
)
if (temp.elements() > 0):
S1[temp] = 0.5 * (3 / 2 - af.abs(
(positions_x_new_5x[temp] - x_grid[index_list[temp]]) / dx))**2
# Computing DS
DS = S1 - S0
W = af.data.moddims(DS, number_of_electrons, 5)
# Computing weights
Jx[:, 0] = -1 * charge_electron * (dx / dt) * W[:, 0].copy()
Jx[:, 1] = Jx[:, 0] + (-1 * charge_electron * (dx / dt) * W[:, 1].copy())
Jx[:, 2] = Jx[:, 1] + (-1 * charge_electron * (dx / dt) * W[:, 2].copy())
Jx[:, 3] = Jx[:, 2] + (-1 * charge_electron * (dx / dt) * W[:, 3].copy())
Jx[:, 4] = Jx[:, 3] + (-1 * charge_electron * (dx / dt) * W[:, 4].copy())
Jx = (1 / dx) * Jx
# Flattening the current matrix
currents = af.flat(Jx)
# temp = af.where(af.abs(currents) > 0)
# if(temp.elements()>0):
# index_list = index_list[temp].copy()
# currents = currents[temp].copy()
af.eval(index_list)
af.eval(currents)
return index_list, currents
def TSC_charge_deposition(charge_electron, no_of_electrons, positions_x,\
x_grid, ghost_cells,length_domain_x, dx):
'''
function cloud_charge_deposition(charge, zone_x, frac_x, x_grid, dx)
-----------------------------------------------------------------------
Input variables: charge, zone_x, frac_x, x_grid, dx
charge: This is a scalar denoting the charge of the macro particle in the PIC code.
zone_x: This is an array of size (number of electrons/macro particles) containing the indices
of the left corners of the grid cells containing the respective particles
frac_x: This is an array of size (number of electrons/macro particles) containing the fractional values of
the positions of particles in their respective grid cells. This is used for linear interpolation
x_grid: This is an array denoting the position grid chosen in the PIC simulation
dx: This is the distance between any two consecutive grid nodes of the position grid.
-----------------------------------------------------------------------
returns: rho
rho: This is an array containing the charges deposited at the density grid nodes.
'''
base_indices = af.data.constant(0,
positions_x.elements(),
dtype=af.Dtype.u32)
x_zone = (((af.abs(positions_x - af.sum(x_grid[0]))) / dx).as_type(
af.Dtype.u32))
temp = af.where(
af.abs(positions_x - x_grid[x_zone]) < af.abs(positions_x -
x_grid[x_zone + 1]))
if (temp.elements() > 0):
base_indices[temp] = x_zone[temp]
temp = af.where(
af.abs(positions_x - x_grid[x_zone]) >= af.abs(positions_x -
x_grid[x_zone + 1]))
if (temp.elements() > 0):
base_indices[temp] = (x_zone[temp] + 1).as_type(af.Dtype.u32)
base_indices_minus = (base_indices - 1).as_type(af.Dtype.u32)
base_indices_plus = (base_indices + 1).as_type(af.Dtype.u32)
index_list = af.join(0, base_indices_minus, base_indices,
base_indices_plus)
positions_x_3x = af.join(0, positions_x, positions_x, positions_x)
distance_nodes = x_grid[index_list]
W = 0 * positions_x_3x.copy()
temp = af.where(af.abs(distance_nodes - positions_x_3x) < dx / 2)
if (temp.elements() > 0):
W[temp] = 0.75 - (af.abs(distance_nodes[temp] - positions_x_3x[temp]) /
dx)**2
temp = af.where((af.abs(distance_nodes - positions_x_3x) >= dx/2 )\
* (af.abs(distance_nodes - positions_x_3x) < (1.5 * dx) )\
)
if (temp.elements() > 0):
W[temp] = 0.5 * (
1.5 -
(af.abs(distance_nodes[temp] - positions_x_3x[temp]) / dx))**2
charges = (charge_electron * W) / dx
# Depositing currents using numpy histogram
input_indices = index_list
elements = x_grid.elements()
rho, temp = np.histogram(input_indices,
bins=elements,
range=(0, elements),
weights=charges)
rho = af.to_array(rho)
af.eval(rho)
return rho
def indices_and_currents_NGP(charge_electron, positions_x, velocity_x, dt,
x_grid, dx, ghost_cells):
positions_x_new = positions_x + velocity_x * dt
number_of_electrons = positions_x.elements()
base_indices = af.data.constant(0,
positions_x.elements(),
dtype=af.Dtype.u32)
Jx = af.data.constant(0, positions_x.elements(), 3, dtype=af.Dtype.f64)
x_zone = (((af.abs(positions_x - af.sum(x_grid[0]))) / dx).as_type(
af.Dtype.u32))
temp = af.where(
af.abs(positions_x - x_grid[x_zone]) < af.abs(positions_x -
x_grid[x_zone + 1]))
if (temp.elements() > 0):
base_indices[temp] = x_zone[temp]
temp = af.where(
af.abs(positions_x - x_grid[x_zone]) >= af.abs(positions_x -
x_grid[x_zone + 1]))
if (temp.elements() > 0):
base_indices[temp] = (x_zone[temp] + 1).as_type(af.Dtype.u32)
S0 = af.data.constant(0, 3 * positions_x.elements(), dtype=af.Dtype.f64)
S1 = S0.copy()
base_indices_plus = (base_indices + 1).as_type(af.Dtype.u32)
base_indices_minus = (base_indices - 1).as_type(af.Dtype.u32)
index_list = af.join(0, base_indices_minus, base_indices,
base_indices_plus)
# Computing S(base_indices - 1, base_indices, base_indices + 1)
positions_x_3x = af.join(0, positions_x, positions_x, positions_x)
positions_x_new_3x = af.join(0, positions_x_new, positions_x_new,
positions_x_new)
temp = af.where(af.abs(positions_x_3x - x_grid[index_list]) < dx / 2)
if (temp.elements() > 0):
S0[temp] = 1
temp = af.where(af.abs(positions_x_new_3x - x_grid[index_list]) < dx / 2)
if (temp.elements() > 0):
S1[temp] = 1
# Computing DS
DS = S1 - S0
DS = af.data.moddims(DS, number_of_electrons, 3)
# Computing weights
W = DS.copy()
Jx[:, 0] = -1 * charge_electron * velocity_x * W[:, 0].copy()
Jx[:, 1] = Jx[:, 0] + (-1 * charge_electron * velocity_x * W[:, 1].copy())
Jx[:, 2] = Jx[:, 1] + (-1 * charge_electron * velocity_x * W[:, 2].copy())
# Flattening the current matrix
currents = af.flat(Jx)
af.eval(index_list)
af.eval(currents)
temp = af.where(af.abs(currents) > 0)
index_list = index_list[temp].copy()
currents = currents[temp].copy()
af.eval(index_list)
af.eval(currents)
return index_list, currents