def blendo(self,
files,
indir,
outdir,
fname,
error=0.05): # default error is 5%
# files is a dictionary. Key = asset name, Value=[filename,fraction] fraction is decimal 0..1
# see trial in __main__ below
# error is the error term as a %, derived from te
self.frame.SetStatusText('Blending files - please wait.')
filenms = [i[0] for i in files]
pcts = [i[1] for i in files]
returns = [loader(fn, indir, self.filetype) for fn in filenms]
balret = [ret * pct for ret, pct in zip(returns, pcts)]
sumbalret = sum(balret)
sds = error * abs(sumbalret)
e1 = np.random.normal(sumbalret, sds)
e2 = pd.DataFrame(e1)
errs = round(e2, 6)
errs.columns = sumbalret.columns
balgro = 1 + sumbalret
errbalgro = 1 + errs
balfundval = np.cumproduct(balgro, axis=1)
errbalfundval = np.cumproduct(errbalgro, axis=1)
cols_needed = [i for i in balfundval.columns if i % 12 == 0]
outputbalfund = round(balfundval[cols_needed], 6)
#outputbalfund.insert(0,-1,1)
#outputbalfund.to_csv(outdir + '//' + fname +'raw.csv', header=False, index=False)
outputerrbal = round(errbalfundval[cols_needed], 6)
#outputerrbal.insert(0,-1,1)
outputerrbal.to_csv(outdir + '//' + fname + '.csv',
header=False,
index=False)
def pnl_plotter(df, price, stock_code):
plt_df = pd.DataFrame()
underlying_df = price['close']
underlying_df.index = price['date_time']
underlying_df = underlying_df.pct_change().dropna()
plt_df['underlying'] = np.cumproduct(underlying_df + 1)
plt_df['strategy'] = np.cumproduct(df['pnl'] + 1)
plt_df['strategy'].ffill(inplace=True)
plt_df['strategy'] = plt_df['strategy'].fillna(1)
plt.plot(plt_df)
plt.title(stock_code)
plt.show()
return (plt_df)
def generate_c_code(self, **kwargs):
TEMPALTE_IDENTITY_FUNC = cleandoc('''
void {op_func_name}(void *op_param, {t} Input{InputDims}, {t} Output{OutputDims}, void *inputs_params, void* outputs_params){{
memcpy(Output, Input, sizeof({t}) * {cumdim});
}}
''')
res = ''
res += self.get_c_param_type() # call only once
res += '\n\n\n'
# constant function
mapping = {}
mapping.update({'op_func_name': self.get_func_name()})
mapping.update({'t': data_type.np2c(self.input_tensor_dtypes[0])})
mapping.update(
{'cumdim': np.cumproduct(self.input_tensor_shapes[0])[-1]})
mapping.update({'Input': self.input_tensor_names[0]})
mapping.update({'Output': self.output_tensor_names[0]})
mapping.update({
'InputDims':
c_helper.generate_dim_bracket(self.input_tensor_shapes[0])
})
mapping.update({
'OutputDims':
c_helper.generate_dim_bracket(self.input_tensor_shapes[0])
})
res += TEMPALTE_IDENTITY_FUNC.format(**mapping)
return res
def calculate_ion_populations(self, phis):
"""
Calculate the ionization balance
.. math::
N(X) = N_1 + N_2 + N_3 + \\dots
N(X) = (N_2/N_1) \\times N_1 + (N3/N2) \\times (N_2/N_1) \\times N_1 + \\dots
N(X) = N_1(1 + N_2/N_1 + (N_3/N_2) \\times (N_2/N_1) + \\dots
N(X) = N_1(1+ \\Phi_{i,j}/N_e + \\Phi_{i, j}/N_e \\times \\Phi_{i, j+1}/N_e + \\dots)
"""
#TODO see if self.ion_populations is None is needed (first class should be enough)
if not hasattr(self, 'ion_populations') or self.ion_populations is None:
self.ion_populations = pd.Series(index=self.partition_functions.index.copy())
for atomic_number, groups in phis.groupby(level='atomic_number'):
current_phis = groups.values / self.electron_density
phis_product = np.cumproduct(current_phis)
neutral_atom_density = self.number_density.ix[atomic_number] / (1 + np.sum(phis_product))
ion_densities = [neutral_atom_density] + list(neutral_atom_density * phis_product)
self.ion_populations.ix[atomic_number] = ion_densities
def __init__(self, upsample_scales, pad):
super(UpsampleNet, self).__init__()
self.upsample_scales = upsample_scales
self.pad = pad
self.total_scale = np.cumproduct(self.upsample_scales)[-1]
self.indent = self.pad * self.total_scale
def calculate_ion_populations(self, phis, ion_zero_threshold=1e-20):
"""
Calculate the ionization balance
.. math::
N(X) = N_1 + N_2 + N_3 + \\dots
N(X) = (N_2/N_1) \\times N_1 + (N3/N2) \\times (N_2/N_1) \\times N_1 + \\dots
N(X) = N_1(1 + N_2/N_1 + (N_3/N_2) \\times (N_2/N_1) + \\dots
N(X) = N_1(1+ \\Phi_{i,j}/N_e + \\Phi_{i, j}/N_e \\times \\Phi_{i, j+1}/N_e + \\dots)
"""
#TODO see if self.ion_populations is None is needed (first class should be enough)
if not hasattr(self, 'ion_populations'):
self.ion_populations = pd.DataFrame(index=self.partition_functions.index.copy(),
columns=np.arange(len(self.t_rads)), dtype=np.float64)
for atomic_number, groups in phis.groupby(level='atomic_number'):
current_phis = (groups / self.electron_densities).replace(np.nan, 0.0).values
phis_product = np.cumproduct(current_phis, axis=0)
neutral_atom_density = self.number_densities.ix[atomic_number] / (1 + np.sum(phis_product, axis=0))
self.ion_populations.ix[atomic_number].values[0] = neutral_atom_density.values
self.ion_populations.ix[atomic_number].values[1:] = neutral_atom_density.values * phis_product
self.ion_populations[self.ion_populations < ion_zero_threshold] = 0.0
def _combine(df, groupers, sort, row_limit=None):
for grouper in groupers:
if isinstance(grouper, Binner):
raise NotImplementedError(
'Cannot combined Binner with other groupers yet')
groupers = groupers.copy()
max_count_64bit = 2**63 - 1
first = groupers.pop(0)
combine_now = [first]
combine_later = []
counts = [first.N]
# when does the cartesian product overflow 64 bits?
next = groupers.pop(0)
while (product(counts) * next.N < max_count_64bit):
counts.append(next.N)
combine_now.append(next)
if groupers:
next = groupers.pop(0)
else:
next = None
break
counts.append(1)
# decreasing [40, 4, 1] for 2 groupers (N=10 and N=4)
cumulative_counts = np.cumproduct(counts[::-1], dtype='i8').tolist()[::-1]
assert len(combine_now) >= 2
combine_later = ([next] if next else []) + groupers
binby_expressions = [df[k.binby_expression] for k in combine_now]
for i in range(0, len(binby_expressions)):
binby_expression = binby_expressions[i]
dtype = vaex.utils.required_dtype_for_max(cumulative_counts[i])
binby_expression = binby_expression.astype(str(dtype))
if isinstance(combine_now[i],
GrouperCategory) and combine_now[i].min_value != 0:
binby_expression -= combine_now[i].min_value
if cumulative_counts[i + 1] != 1:
binby_expression = binby_expression * cumulative_counts[i + 1]
binby_expressions[i] = binby_expression
expression = reduce(operator.add, binby_expressions)
grouper = GrouperCombined(expression,
df,
multipliers=cumulative_counts[1:],
parents=combine_now,
sort=sort,
row_limit=row_limit)
if combine_later:
@vaex.delayed
def combine(_ignore):
# recursively add more of the groupers (because of 64 bit overflow)
# return 1
grouper._create_binner(df)
new_grouper = _combine(df, [grouper] + combine_later, sort=sort)
return new_grouper
return combine(grouper._promise)
return grouper._promise.then(lambda x: grouper)
def pis(self):
#stick breaking procedure
betas = np.random.beta(1, self.alpha, self.k)
# compute prod(1- beta)_i^n-1
cum_prods = np.append(1, np.cumproduct(1 - betas[:-1]))
prods = betas * cum_prods
return prods / prods.sum()
def __init__(
self,
feat_dims,
upsample_scales,
compute_dims,
num_res_blocks,
res_out_dims,
pad,
use_aux_net,
):
super().__init__()
self.total_scale = np.cumproduct(upsample_scales)[-1]
self.indent = pad * self.total_scale
self.use_aux_net = use_aux_net
if use_aux_net:
self.resnet = MelResNet(num_res_blocks, feat_dims, compute_dims,
res_out_dims, pad)
self.resnet_stretch = Stretch2d(self.total_scale, 1)
self.up_layers = nn.ModuleList()
for scale in upsample_scales:
k_size = (1, scale * 2 + 1)
padding = (0, scale)
stretch = Stretch2d(scale, 1)
conv = nn.Conv2d(1,
1,
kernel_size=k_size,
padding=padding,
bias=False)
conv.weight.data.fill_(1.0 / k_size[1])
self.up_layers.append(stretch)
self.up_layers.append(conv)
def __init__(self,
feat_dims,
upsample_scales=[4, 4, 10],
compute_dims=128,
res_blocks=10,
res_out_dims=128,
pad=2):
super().__init__()
self.num_outputs = res_out_dims
total_scale = np.cumproduct(upsample_scales)[-1]
self.indent = pad * total_scale
self.resnet = MelResNet(res_blocks, feat_dims, compute_dims,
res_out_dims, pad)
self.resnet_stretch = Stretch2d(total_scale, 1)
self.up_layers = nn.ModuleList()
for scale in upsample_scales:
k_size = (1, scale * 2 + 1)
padding = (0, scale)
stretch = Stretch2d(scale, 1)
conv = nn.Conv2d(1,
1,
kernel_size=k_size,
padding=padding,
bias=False)
conv.weight.data.fill_(1. / k_size[1])
self.up_layers.append(stretch)
self.up_layers.append(conv)
def calculate_ion_populations(self, phis):
"""
Calculate the ionization balance
.. math::
N(X) = N_1 + N_2 + N_3 + \\dots
N(X) = (N_2/N_1) \\times N_1 + (N3/N2) \\times (N_2/N_1) \\times N_1 + \\dots
N(X) = N_1(1 + N_2/N_1 + (N_3/N_2) \\times (N_2/N_1) + \\dots
N(X) = N_1(1+ \\Phi_{i,j}/N_e + \\Phi_{i, j}/N_e \\times \\Phi_{i, j+1}/N_e + \\dots)
"""
#TODO see if self.ion_populations is None is needed (first class should be enough)
if not hasattr(self,
'ion_populations') or self.ion_populations is None:
self.ion_populations = pd.Series(
index=self.partition_functions.index.copy())
for atomic_number, groups in phis.groupby(level='atomic_number'):
current_phis = groups.values / self.electron_density
phis_product = np.cumproduct(current_phis)
neutral_atom_density = self.number_density.ix[atomic_number] / (
1 + np.sum(phis_product))
ion_densities = [neutral_atom_density] + list(
neutral_atom_density * phis_product)
self.ion_populations.ix[atomic_number] = ion_densities
def prepare_diffusion_vars(self):
"""Prepare for variables used in the diffusion process."""
self.betas = self.get_betas()
self.alphas = 1.0 - self.betas
self.alphas_bar = np.cumproduct(self.alphas, axis=0)
self.alphas_bar_prev = np.append(1.0, self.alphas_bar[:-1])
self.alphas_bar_next = np.append(self.alphas_bar[1:], 0.0)
# calculations for diffusion q(x_t | x_0) and others
self.sqrt_alphas_bar = np.sqrt(self.alphas_bar)
self.sqrt_one_minus_alphas_bar = np.sqrt(1.0 - self.alphas_bar)
self.log_one_minus_alphas_bar = np.log(1.0 - self.alphas_bar)
self.sqrt_recip_alplas_bar = np.sqrt(1.0 / self.alphas_bar)
self.sqrt_recipm1_alphas_bar = np.sqrt(1.0 / self.alphas_bar - 1)
# calculations for posterior q(x_{t-1} | x_t, x_0)
self.tilde_betas_t = self.betas * (1 - self.alphas_bar_prev) / (
1 - self.alphas_bar)
# clip log var for tilde_betas_0 = 0
self.log_tilde_betas_t_clipped = np.log(
np.append(self.tilde_betas_t[1], self.tilde_betas_t[1:]))
self.tilde_mu_t_coef1 = np.sqrt(
self.alphas_bar_prev) / (1 - self.alphas_bar) * self.betas
self.tilde_mu_t_coef2 = np.sqrt(
self.alphas) * (1 - self.alphas_bar_prev) / (1 - self.alphas_bar)
def cartesian_product(X):
"""
Numpy version of itertools.product or pandas.compat.product.
Sometimes faster (for large inputs)...
Examples
--------
>>> cartesian_product([list('ABC'), [1, 2]])
[array(['A', 'A', 'B', 'B', 'C', 'C'], dtype='|S1'),
array([1, 2, 1, 2, 1, 2])]
"""
lenX = np.fromiter((len(x) for x in X), dtype=int)
cumprodX = np.cumproduct(lenX)
a = np.roll(cumprodX, 1)
a[0] = 1
b = cumprodX[-1] / cumprodX
return [
np.tile(np.repeat(np.asarray(com._values_from_object(x)), b[i]),
np.product(a[i])) for i, x in enumerate(X)
]
def calculate_ion_populations(self, phis, ion_zero_threshold=1e-20):
"""
Calculate the ionization balance
.. math::
N(X) = N_1 + N_2 + N_3 + \\dots
N(X) = (N_2/N_1) \\times N_1 + (N3/N2) \\times (N_2/N_1) \\times N_1 + \\dots
N(X) = N_1(1 + N_2/N_1 + (N_3/N_2) \\times (N_2/N_1) + \\dots
N(X) = N_1(1+ \\Phi_{i,j}/N_e + \\Phi_{i, j}/N_e \\times \\Phi_{i, j+1}/N_e + \\dots)
"""
#TODO see if self.ion_populations is None is needed (first class should be enough)
if not hasattr(self, 'ion_populations'):
self.ion_populations = pd.DataFrame(index=self.partition_functions.index.copy(),
columns=np.arange(len(self.t_rads)), dtype=np.float64)
for atomic_number, groups in phis.groupby(level='atomic_number'):
current_phis = (groups / self.electron_densities).replace(np.nan, 0.0).values
phis_product = np.cumproduct(current_phis, axis=0)
neutral_atom_density = self.number_densities.ix[atomic_number] / (1 + np.sum(phis_product, axis=0))
self.ion_populations.ix[atomic_number].values[0] = neutral_atom_density.values
self.ion_populations.ix[atomic_number].values[1:] = neutral_atom_density.values * phis_product
self.ion_populations[self.ion_populations < ion_zero_threshold] = 0.0
def sample_episode(env, action_policy, gamma):
lists_a, lists_s, lists_s_ne, lists_r = [], [], [], []
done = False
s = env.reset()
step_counter = 0
while not done:
a = action_policy(torch.tensor(s.reshape(1, -1), dtype=torch.float))
s_ne, r, done, _ = env.step(a)
lists_a.append(a), lists_s.append(s), lists_s_ne.append(
s_ne), lists_r.append(r)
step_counter += 1
s = s_ne
assert step_counter == 5000 # not sure if the stepsize is fixed of 5000
actions = np.array(lists_a)
states = np.array(lists_s)
rewards = np.array(lists_r)
gamma_array = np.ones_like(rewards) * gamma
gamma_array = np.cumproduct(gamma_array)
R_fun = gamma_array * rewards
print("mean_reward in this sampling ", np.sum(rewards) / step_counter)
return states, actions.reshape(-1, 1), R_fun.reshape(
-1, 1), np.sum(rewards) / step_counter
def gradient(x, t, p, log_p):
'''
Gradient of the objective with decision variables x and scenario t. p is the
detection probabilities and log_p is log(1 - p) (elementwise).
Computations are performed via vectorized numpy operations for greater efficiency.
'''
import numpy as np
order = np.argsort(t)
x = x[order]
p = p[order]
log_p = log_p[order]
t = t[order]
p_fail = ((1 - p)**x)
cum_prob_fail = np.cumproduct(p_fail)
p_fail_offset = np.zeros((len(x) + 1))
p_fail_offset[0] = 1
p_fail_offset[1:] = cum_prob_fail
cum_sum = np.zeros((len(x)))
cum_sum[-1] = 0
intermediate = (p_fail_offset[:-1] * (1 - p_fail) * t)[1:]
intermediate = intermediate[::-1]
intermediate = np.cumsum(intermediate)
intermediate = intermediate[::-1]
cum_sum[:-1] = intermediate
ordered_grad = t * log_p * cum_prob_fail - log_p * cum_sum - log_p * cum_prob_fail[
-1] * t.max()
ordered_grad[np.abs(ordered_grad) < 0.00001] = 0
new_perm = np.zeros((len(order)), dtype=np.int)
for i, val in enumerate(order):
new_perm[val] = i
return ordered_grad[new_perm]
def __init__(self, shape_, order='C', **keywords):
shape_ = tointtuple(shape_)
ndim = len(shape_)
if ndim == 1:
raise NotImplementedError('ndim == 1 is not implemented.')
if order.upper() not in ('C', 'F'):
raise ValueError("Invalid order '{0}'. Expected order is 'C' or 'F"
"'".format(order))
order = order.upper()
Operator.__init__(self, **keywords)
self.shape_ = shape_
self.order = order
self.ndim = ndim
if order == 'C':
self.coefs = np.cumproduct((1,) + shape_[:0:-1])[::-1]
elif order == 'F':
self.coefs = np.cumproduct((1,) + shape_[:-1])
def __init__(self, shape_, order='C', **keywords):
shape_ = tointtuple(shape_)
ndim = len(shape_)
if ndim == 1:
raise NotImplementedError('ndim == 1 is not implemented.')
if order.upper() not in ('C', 'F'):
raise ValueError("Invalid order '{0}'. Expected order is 'C' or 'F"
"'".format(order))
order = order.upper()
Operator.__init__(self, **keywords)
self.shape_ = shape_
self.order = order
self.ndim = ndim
if order == 'C':
self.coefs = np.cumproduct((1, ) + shape_[:0:-1])[::-1]
elif order == 'F':
self.coefs = np.cumproduct((1, ) + shape_[:-1])
def __init__(self, rnn_dims, fc_dims, mode, mulaw, pad, use_aux_net, use_upsample_net, upsample_factors,
feat_dims, compute_dims, res_out_dims, res_blocks,
hop_length, sample_rate):
super().__init__()
self.mode = mode
self.mulaw = mulaw
self.pad = pad
self.use_upsample_net = use_upsample_net
self.use_aux_net = use_aux_net
if type(self.mode) is int:
self.n_classes = 2 ** self.mode
elif self.mode == 'mold':
self.n_classes = 3 * 10
elif self.mode == 'gauss':
self.n_classes = 2
else:
raise RuntimeError(" > Unknown training mode")
self.rnn_dims = rnn_dims
self.aux_dims = res_out_dims // 4
self.hop_length = hop_length
self.sample_rate = sample_rate
print(np.cumproduct(upsample_factors))
if self.use_upsample_net:
print(np.cumproduct(upsample_factors)[-1], self.hop_length)
assert np.cumproduct(upsample_factors)[-1] == self.hop_length, " [!] upsample scales needs to be equal to hop_length"
self.upsample = UpsampleNetwork(feat_dims, upsample_factors, compute_dims,
res_blocks, res_out_dims, pad, use_aux_net)
else:
self.upsample = Upsample(hop_length, pad, res_blocks, feat_dims, compute_dims, res_out_dims, use_aux_net)
if self.use_aux_net:
self.I = nn.Linear(feat_dims + self.aux_dims + 1, rnn_dims)
self.rnn1 = nn.GRU(rnn_dims, rnn_dims, batch_first=True)
self.rnn2 = nn.GRU(rnn_dims + self.aux_dims, rnn_dims, batch_first=True)
self.fc1 = nn.Linear(rnn_dims + self.aux_dims, fc_dims)
self.fc2 = nn.Linear(fc_dims + self.aux_dims, fc_dims)
self.fc3 = nn.Linear(fc_dims, self.n_classes)
else:
self.I = nn.Linear(feat_dims + 1, rnn_dims)
self.rnn1 = nn.GRU(rnn_dims, rnn_dims, batch_first=True)
self.rnn2 = nn.GRU(rnn_dims, rnn_dims, batch_first=True)
self.fc1 = nn.Linear(rnn_dims, fc_dims)
self.fc2 = nn.Linear(fc_dims, fc_dims)
self.fc3 = nn.Linear(fc_dims, self.n_classes)
def shift(array, n, axis=0):
"""
Shift array elements inplace along a given axis.
Elements that are shifted beyond the last position are not re-introduced
at the first.
Parameters
----------
array : float array
Input array to be modified
n : integer number or array
The number of places by which elements are shifted. If it is an array,
specific offsets are applied along the first dimensions.
axis : int, optional
The axis along which elements are shifted. By default, it is the first
axis.
Examples
--------
>>> a = ones(8)
>>> shift(a, 3); a
array([0., 0., 0., 1., 1., 1., 1., 1., 1.])
>>> a = array([[1.,1.,1.,1.],[2.,2.,2.,2.]])
>>> shift(a, [1,-1], axis=1); a
array([[0., 1., 1., 1.],
[2., 2., 2., 0.]])
"""
if not isinstance(array, np.ndarray):
raise TypeError('Input array is not an ndarray.')
if array.dtype != var.FLOAT_DTYPE:
raise TypeError('The data type of the input array is not ' + \
str(var.FLOAT_DTYPE.name) + '.')
rank = array.ndim
n = np.array(n, ndmin=1, dtype='int32').ravel()
if axis < 0:
axis = rank + axis
if axis == 0 and n.size > 1 or n.size != 1 and n.size not in \
np.cumproduct(array.shape[0:axis]):
raise ValueError('The offset size is incompatible with the first dime' \
'nsions of the array')
if rank == 0:
array.shape = (1,)
array[:] = 0
array.shape = ()
else:
tmf.shift(array.ravel(), rank-axis, np.asarray(array.T.shape), n)
def __init__(self,
time_size,
space_size,
dim,
beta,
link_type,
num_samples=None,
rand=False):
"""Initialization for GaugeLattice object.
Args:
time_size (int): Temporal extent of lattice.
space_size (int): Spatial extent of lattice.
dim (int): Dimensionality
beta (float): Inverse coupling constant.
link_type (str): String representing the type of gauge group for
the link variables. Must be either 'U1', 'SU2', or 'SU3'
num_samples (int): Number of sample lattices to use.
rand (bool): Flag specifying if lattice should be initialized
randomly or uniformly.
"""
assert link_type.upper() in [
'U1', 'SU2', 'SU3'
], ("Invalid link_type. Possible values: U1', 'SU2', 'SU3'")
self.time_size = time_size
self.space_size = space_size
self.dim = dim
self.beta = beta
self.link_type = link_type
self.link_shape = None
self._init_lattice(link_type, rand)
self.samples = None
self.num_sites = np.cumproduct(self.site_idxs)[-1]
self.num_links = int(self.dim * self.num_sites)
self.num_plaquettes = self.time_size * self.space_size
self.bases = np.eye(dim, dtype=np.int)
if self.link_type == 'U1':
self.plaquette_operator = self.plaquette_operator_u1
self._action_op = self._action_op_u1
else:
self.plaquette_operator = self.plaquette_operator_suN
self.action_op = self._action_op_suN
if num_samples is not None:
# Create `num_samples` randomized instances of links array
self.num_samples = num_samples
self.samples = self.get_links_samples(num_samples,
rand=rand,
link_type=self.link_type)
self.samples[0] = self.links
def shift(array, n, axis=0):
"""
Shift array elements inplace along a given axis.
Elements that are shifted beyond the last position are not re-introduced
at the first.
Parameters
----------
array : float array
Input array to be modified
n : integer number or array
The number of places by which elements are shifted. If it is an array,
specific offsets are applied along the first dimensions.
axis : int, optional
The axis along which elements are shifted. By default, it is the first
axis.
Examples
--------
>>> a = ones(8)
>>> shift(a, 3); a
array([0., 0., 0., 1., 1., 1., 1., 1., 1.])
>>> a = array([[1.,1.,1.,1.],[2.,2.,2.,2.]])
>>> shift(a, [1,-1], axis=1); a
array([[0., 1., 1., 1.],
[2., 2., 2., 0.]])
"""
if not isinstance(array, np.ndarray):
raise TypeError('Input array is not an ndarray.')
if array.dtype != var.FLOAT_DTYPE:
raise TypeError('The data type of the input array is not ' + \
str(var.FLOAT_DTYPE.name) + '.')
rank = array.ndim
n = np.array(n, ndmin=1, dtype='int32').ravel()
if axis < 0:
axis = rank + axis
if axis == 0 and n.size > 1 or n.size != 1 and n.size not in \
np.cumproduct(array.shape[0:axis]):
raise ValueError('The offset size is incompatible with the first dime' \
'nsions of the array')
if rank == 0:
array.shape = (1, )
array[:] = 0
array.shape = ()
else:
tmf.shift(array.ravel(), rank - axis, np.asarray(array.T.shape), n)
def __init__(self, res_blocks, upsample_scales, compute_dims, output_dims,
pad):
super(UpsampleNet, self).__init__()
self.res_blocks = res_blocks
self.upsample_scales = upsample_scales
self.compute_dims = compute_dims
self.output_dims = output_dims
self.pad = pad
self.total_scale = np.cumproduct(self.upsample_scales)[-1]
self.indent = self.pad * self.total_scale
def cartesian_product(X) -> list[np.ndarray]:
"""
Numpy version of itertools.product.
Sometimes faster (for large inputs)...
Parameters
----------
X : list-like of list-likes
Returns
-------
product : list of ndarrays
Examples
--------
>>> cartesian_product([list('ABC'), [1, 2]])
[array(['A', 'A', 'B', 'B', 'C', 'C'], dtype='<U1'), array([1, 2, 1, 2, 1, 2])]
See Also
--------
itertools.product : Cartesian product of input iterables. Equivalent to
nested for-loops.
"""
msg = "Input must be a list-like of list-likes"
if not is_list_like(X):
raise TypeError(msg)
for x in X:
if not is_list_like(x):
raise TypeError(msg)
if len(X) == 0:
return []
lenX = np.fromiter((len(x) for x in X), dtype=np.intp)
cumprodX = np.cumproduct(lenX)
if np.any(cumprodX < 0):
raise ValueError("Product space too large to allocate arrays!")
a = np.roll(cumprodX, 1)
a[0] = 1
if cumprodX[-1] != 0:
b = cumprodX[-1] / cumprodX
else:
# if any factor is empty, the cartesian product is empty
b = np.zeros_like(cumprodX)
return [
tile_compat(np.repeat(x, b[i]), np.product(a[i]))
for i, x in enumerate(X)
]
def alpha(self, pH):
'''Return the fraction of each species at a given pH.
Parameters
----------
pH : int, float, or Numpy Array
These are the pH value(s) over which the fraction should be
returned.
Returns
-------
Numpy NDArray
These are the fractional concentrations at any given pH. They are
sorted from most acidic species to least acidic species. If a
NDArray of pH values is provided, then a 2D array will be
returned. In this case, each row represents the speciation for
each given pH.
'''
# If the given pH is not a list/array, be sure to convert it to one
# for future calcs.
if isinstance(pH, (int, float)):
pH = [
pH,
]
pH = np.array(pH, dtype=float)
# Calculate the concentration of H3O+. If multiple pH values are
# given, then it is best to construct a two dimensional array of
# concentrations.
h3o = 10.**(-pH)
if len(h3o) > 1:
h3o = np.repeat(h3o.reshape(-1, 1), len(self._Ka_temp), axis=1)
# These are the powers that the H3O+ concentrations will be raised.
power = np.arange(len(self._Ka_temp))
# Calculate the H3O+ concentrations raised to the powers calculated
# above (in reverse order).
h3o_pow = h3o**(power[::-1])
# Calculate a cumulative product of the Ka values. The first value
# must be 1.0, which is why _Ka_temp is used instead of Ka.
Ka_prod = np.cumproduct(self._Ka_temp)
# Multiply the H3O**power values times the cumulative Ka product.
h3o_Ka = h3o_pow * Ka_prod
# Return the alpha values. The return signature will differ is the
# shape of the H3O array was 2-dimensional.
if len(h3o.shape) > 1:
den = h3o_Ka.sum(axis=1)
return h3o_Ka / den.reshape(-1, 1)
else:
den = h3o_Ka.sum()
return h3o_Ka / den
def index_to_assignment(
I, D): # i is an integer and D is a python list eg [2, 2, 2]
g = list(D[0:len(D) - 1])
g.insert(0, 1)
result = np.float32(
np.mod(
np.floor(
np.divide(npm.repmat(I - 1, 1, len(D)),
npm.repmat(np.cumproduct(g), 1, 1))),
npm.repmat(D, 1, 1)))
return result.flatten()
def reshape(self, shape):
try:
shape = tuple(shape)
except TypeError:
shape = (shape, )
if np.product(shape) != len(self):
raise InvalidArgument(
"Reshape failed: New shape %s has different length to "
"existing shape %s" % (str(shape), str(self.shape)))
bools = self.as_boolarray().reshape(-1)
buf = np.packbits(bools)
strides = [1] + list(np.cumproduct(shape[-1:0:-1]))
return BitArray(buf, 0, shape, strides)
def f(x, t, p):
'''
Objective value to decision x in scenario with reachability times t and detection probabilities p
'''
import numpy as np
order = np.argsort(t)
prob_fail = np.zeros((len(x) + 1))
prob_fail[0] = 1
prob_fail[1:] = ((1 - p)**x)[order]
cum_prob_fail = np.cumproduct(prob_fail)
prob_succeed = (1 - prob_fail)[1:]
return t.max() - (cum_prob_fail[:-1] * prob_succeed *
t[order]).sum() - cum_prob_fail[-1] * t.max()
def __init__(self, description, dimension=None, offset=None, strides=None):
dimension = [1] if dimension is None else dimension[:] # copy
strides = (np.cumproduct(np.hstack((1, dimension)))[:-1]
if strides is None else strides)
self._nodes = copy.deepcopy(Detector._nodes)
self._nodes['dimension']['shape'] = [len(dimension)]
self._nodes['strides']['shape'] = [len(strides)]
self._nodes['counts']['shape'] = dimension
self._nodes['roiMask']['shape'] = dimension
self._nodes['sliceCounts']['shape'] = dimension
self.dimension = dimension
self.strides = strides
self.offset = 0 if offset is None else offset
self.strides = (np.cumproduct(np.hstack((1, dimension)))[:-1].tolist()
if strides is None else strides)
self.roiMask = np.ones(dimension) #.tolist()
self.roiShape = ["name", "unknown"]
self.liveROI = 0.
self.counts = np.zeros(dimension, 'int32') #.tolist()
self.sliceCounts = np.zeros(dimension, 'int32') #.tolist()
self._description = description
def generate_encode_macro(name, shape):
multipliers = [1] + list(np.cumproduct(shape[::-1]))[:-1]
multipliers = [Constant(i) for i in reversed(multipliers)]
params = [SymbolRef(name='x{}'.format(dim)) for dim in range(len(shape))]
variables = []
for var in params:
cp = var.copy()
cp._force_parentheses = True
variables.append(Cast(ctypes.c_long(), cp))
products = [Mul(mult, var) for mult, var in zip(multipliers, variables)]
total = functools.reduce(Add, products)
return CppDefine(name=name, params=params, body=total)
def cartesian_product(X):
"""
Numpy version of itertools.product or pandas.compat.product.
Sometimes faster (for large inputs)...
Parameters
----------
X : list-like of list-likes
Returns
-------
product : list of ndarrays
Examples
--------
>>> cartesian_product([list('ABC'), [1, 2]])
[array(['A', 'A', 'B', 'B', 'C', 'C'], dtype='|S1'),
array([1, 2, 1, 2, 1, 2])]
See also
--------
itertools.product : Cartesian product of input iterables. Equivalent to
nested for-loops.
pandas.compat.product : An alias for itertools.product.
"""
msg = "Input must be a list-like of list-likes"
if not is_list_like(X):
raise TypeError(msg)
for x in X:
if not is_list_like(x):
raise TypeError(msg)
if len(X) == 0:
return []
lenX = np.fromiter((len(x) for x in X), dtype=int)
cumprodX = np.cumproduct(lenX)
a = np.roll(cumprodX, 1)
a[0] = 1
if cumprodX[-1] != 0:
b = cumprodX[-1] / cumprodX
else:
# if any factor is empty, the cartesian product is empty
b = np.zeros_like(cumprodX)
return [
np.tile(np.repeat(np.asarray(com._values_from_object(x)), b[i]),
np.product(a[i])) for i, x in enumerate(X)
]
def save_portfolio_metrics(portfolios, portfolio_name, period_ends, prices, p_value, p_weights, p_holdings, path=None):
rebalance_qtys = (p_weights.ix[period_ends] / prices.ix[period_ends]) * p_value.ix[period_ends]
# p_holdings = rebalance_qtys.align(prices)[0].shift(1).ffill().fillna(0)
transactions = p_holdings - p_holdings.shift(1).fillna(0)
transactions = transactions[transactions.sum(1) != 0]
p_returns = p_value.pct_change(periods=1)
p_index = np.cumproduct(1 + p_returns)
m_rets = (1 + p_returns).resample("M", how="prod", kind="period") - 1
portfolios[portfolio_name]["equity"] = p_value
portfolios[portfolio_name]["ret"] = p_returns
portfolios[portfolio_name]["cagr"] = compute_cagr(p_value) * 100
portfolios[portfolio_name]["sharpe"] = compute_sharpe(p_value)
portfolios[portfolio_name]["weight"] = p_weights
portfolios[portfolio_name]["transactions"] = transactions
portfolios[portfolio_name]["period_return"] = 100 * (p_value.ix[-1] / p_value[0] - 1)
portfolios[portfolio_name]["avg_monthly_return"] = p_index.resample("BM", how="last").pct_change().mean() * 100
portfolios[portfolio_name]["monthly_return_table"] = Monthly_Return_Table(m_rets)
portfolios[portfolio_name]["drawdowns"] = compute_drawdown(p_value).dropna()
portfolios[portfolio_name]["max_drawdown"] = compute_max_drawdown(p_value) * 100
portfolios[portfolio_name]["max_drawdown_date"] = (
p_value.index[compute_drawdown(p_value) == compute_max_drawdown(p_value)][0].date().isoformat()
)
portfolios[portfolio_name]["avg_drawdown"] = compute_avg_drawdown(p_value) * 100
portfolios[portfolio_name]["calmar"] = compute_calmar(p_value)
portfolios[portfolio_name]["R_squared"] = compute_calmar(p_value)
portfolios[portfolio_name]["DVR"] = compute_DVR(p_value)
portfolios[portfolio_name]["volatility"] = compute_volatility(p_returns)
portfolios[portfolio_name]["VAR"] = compute_var(p_returns)
portfolios[portfolio_name]["CVAR"] = compute_cvar(p_returns)
portfolios[portfolio_name]["rolling_annual_returns"] = pd.rolling_apply(p_returns, 252, np.sum)
portfolios[portfolio_name]["p_holdings"] = p_holdings
portfolios[portfolio_name]["transactions"] = np.round(transactions[transactions.sum(1) != 0], 0)
portfolios[portfolio_name]["share"] = p_holdings
portfolios[portfolio_name]["orders"] = generate_orders(transactions, prices)
portfolios[portfolio_name]["best"] = max(p_returns)
portfolios[portfolio_name]["worst"] = min(p_returns)
portfolios[portfolio_name]["trades"] = len(portfolios[portfolio_name]["orders"])
if path != None:
portfolios[portfolio_name].equity.to_csv(path + portfolio_name + "_equity.csv")
portfolios[portfolio_name].weight.to_csv(path + portfolio_name + "_weight.csv")
portfolios[portfolio_name].share.to_csv(path + portfolio_name + "_share.csv")
portfolios[portfolio_name].transactions.to_csv(path + portfolio_name + "_transactions.csv")
portfolios[portfolio_name].orders.to_csv(path + portfolio_name + "_orders.csv")
return
def alpha(self, pH):
'''Return the fraction of each species at a given pH.
Parameters
----------
pH : int, float, or Numpy Array
These are the pH value(s) over which the fraction should be
returned.
Returns
-------
Numpy NDArray
These are the fractional concentrations at any given pH. They are
sorted from most acidic species to least acidic species. If a
NDArray of pH values is provided, then a 2D array will be
returned. In this case, each row represents the speciation for
each given pH.
'''
# If the given pH is not a list/array, be sure to convert it to one
# for future calcs.
if isinstance(pH, (int, float)):
pH = [pH,]
pH = np.array(pH, dtype=float)
# Calculate the concentration of H3O+. If multiple pH values are
# given, then it is best to construct a two dimensional array of
# concentrations.
h3o = 10.**(-pH)
if len(h3o) > 1:
h3o = np.repeat( h3o.reshape(-1, 1), len(self._Ka_temp), axis=1)
# These are the powers that the H3O+ concentrations will be raised.
power = np.arange(len(self._Ka_temp))
# Calculate the H3O+ concentrations raised to the powers calculated
# above (in reverse order).
h3o_pow = h3o**( power[::-1] )
# Calculate a cumulative product of the Ka values. The first value
# must be 1.0, which is why _Ka_temp is used instead of Ka.
Ka_prod = np.cumproduct(self._Ka_temp)
# Multiply the H3O**power values times the cumulative Ka product.
h3o_Ka = h3o_pow*Ka_prod
# Return the alpha values. The return signature will differ is the
# shape of the H3O array was 2-dimensional.
if len(h3o.shape) > 1:
den = h3o_Ka.sum(axis=1)
return h3o_Ka/den.reshape(-1,1)
else:
den = h3o_Ka.sum()
return h3o_Ka/den
def cartesian_product(X):
"""
Numpy version of itertools.product or pandas.compat.product.
Sometimes faster (for large inputs)...
Parameters
----------
X : list-like of list-likes
Returns
-------
product : list of ndarrays
Examples
--------
>>> cartesian_product([list('ABC'), [1, 2]])
[array(['A', 'A', 'B', 'B', 'C', 'C'], dtype='|S1'),
array([1, 2, 1, 2, 1, 2])]
See also
--------
itertools.product : Cartesian product of input iterables. Equivalent to
nested for-loops.
pandas.compat.product : An alias for itertools.product.
"""
msg = "Input must be a list-like of list-likes"
if not is_list_like(X):
raise TypeError(msg)
for x in X:
if not is_list_like(x):
raise TypeError(msg)
if len(X) == 0:
return []
lenX = np.fromiter((len(x) for x in X), dtype=np.intp)
cumprodX = np.cumproduct(lenX)
a = np.roll(cumprodX, 1)
a[0] = 1
if cumprodX[-1] != 0:
b = cumprodX[-1] / cumprodX
else:
# if any factor is empty, the cartesian product is empty
b = np.zeros_like(cumprodX)
return [np.tile(np.repeat(np.asarray(com.values_from_object(x)), b[i]),
np.product(a[i]))
for i, x in enumerate(X)]
def price_monte_carlo(self, iterations, control = True):
drift = math.exp( (self.rate - 0.5 * self.iv ** 2) * self.T / self.steps )
listAP = np.zeros(iterations)
listGP = np.zeros(iterations)
for i in range(iterations):
np.random.seed(i)
arrZ = np.random.normal(0, 1, self.steps)
arrR = np.exp(self.iv * math.sqrt( self.T / self.steps ) * arrZ) * drift
arrP = self.spot * np.cumproduct( arrR )
arithmeticMean = np.mean( arrP )
if self.option_type == 'call':
arithmeticPayoff = math.exp( -self.rate * self.T ) * max ( [ arithmeticMean - self.strike, 0 ] )
else:
arithmeticPayoff = math.exp( -self.rate * self.T ) * max ( [ self.strike - arithmeticMean, 0 ] )
listAP[i] = arithmeticPayoff
geometricMean = np.exp( 1 / self.steps * sum ( np.log( arrP ) ) )
if self.option_type == 'call':
geometricPayoff = math.exp( -self.rate * self.T ) * max ( [ geometricMean - self.strike, 0 ] )
else:
geometricPayoff = math.exp( -self.rate * self.T ) * max ( [ self.strike - geometricMean, 0 ] )
listGP[i] = geometricPayoff
meanGP = np.mean(listGP)
meanAP = np.mean(listAP)
sdAP = np.std (listAP, ddof = 1)
pxClose = self.price()
if self.option_class == 'geometric':
print("[Geometric Asian from Standard MC: {0:6f}, Closed-form price: {1:6f}]".format(meanGP, pxClose))
return (meanGP, pxClose)
if self.option_class == 'arithmetic':
if control:
covPP = np.mean(listGP * listAP) - np.mean(listGP) * np.mean(listAP)
theta = covPP / np.var(listGP, ddof = 1)
listCV = listAP + theta * (pxClose - listGP)
meanCV = np.mean(listCV)
sdCV = np.std(listCV, ddof = 1)
ciCV = (meanCV - 1.96 * sdCV / math.sqrt(iterations), meanCV + 1.96 * sdCV / math.sqrt(iterations), meanCV, pxClose)
print("Control Variate: [Confidence interval: ({0:6f}, {1:6f}), Mean Price: {2:6f}, Closed-form geometric price: {3:6f}]".format(*ciCV))
return ciCV
else:
ciAP = (meanAP - 1.96 * sdAP / math.sqrt(iterations), meanAP + 1.96 * sdAP / math.sqrt(iterations), meanAP, pxClose)
print("No Control: [Confidence interval: ({0:6f}, {1:6f}), Mean Price: {2:6f}, Closed-form geometric price: {3:6f}]".format(*ciAP))
return ciAP
def save_portfolio_metrics (portfolios, portfolio_name, period_ends, prices, \
p_value, p_weights, p_holdings, path=None) :
rebalance_qtys = (p_weights.ix[period_ends] / prices.ix[period_ends]) * p_value.ix[period_ends]
#p_holdings = rebalance_qtys.align(prices)[0].shift(1).ffill().fillna(0)
transactions = (p_holdings - p_holdings.shift(1).fillna(0))
transactions = transactions[transactions.sum(1) != 0]
p_returns = p_value.pct_change(periods=1)
p_index = np.cumproduct(1 + p_returns)
m_rets = (1 + p_returns).resample('M', how='prod', kind='period') - 1
portfolios[portfolio_name]['equity'] = p_value
portfolios[portfolio_name]['ret'] = p_returns
portfolios[portfolio_name]['cagr'] = compute_cagr(p_value) * 100
portfolios[portfolio_name]['sharpe'] = compute_sharpe(p_value)
portfolios[portfolio_name]['weight'] = p_weights
portfolios[portfolio_name]['transactions'] = transactions
portfolios[portfolio_name]['period_return'] = 100 * (p_value.ix[-1] / p_value[0] - 1)
portfolios[portfolio_name]['avg_monthly_return'] = p_index.resample('BM', how='last').pct_change().mean() * 100
portfolios[portfolio_name]['monthly_return_table'] = Monthly_Return_Table(m_rets)
portfolios[portfolio_name]['drawdowns'] = compute_drawdown(p_value).dropna()
portfolios[portfolio_name]['max_drawdown'] = compute_max_drawdown(p_value) * 100
portfolios[portfolio_name]['max_drawdown_date'] = p_value.index[compute_drawdown(p_value)==compute_max_drawdown(p_value)][0].date().isoformat()
portfolios[portfolio_name]['avg_drawdown'] = compute_avg_drawdown(p_value) * 100
portfolios[portfolio_name]['calmar'] = compute_calmar(p_value)
portfolios[portfolio_name]['R_squared'] = compute_calmar(p_value)
portfolios[portfolio_name]['DVR'] = compute_DVR(p_value)
portfolios[portfolio_name]['volatility'] = compute_volatility(p_returns)
portfolios[portfolio_name]['VAR'] = compute_var(p_returns)
portfolios[portfolio_name]['CVAR'] = compute_cvar(p_returns)
portfolios[portfolio_name]['rolling_annual_returns'] = pd.rolling_apply(p_returns, 252, np.sum)
portfolios[portfolio_name]['p_holdings'] = p_holdings
portfolios[portfolio_name]['transactions'] = np.round(transactions[transactions.sum(1)!=0], 0)
portfolios[portfolio_name]['share'] = p_holdings
portfolios[portfolio_name]['orders'] = generate_orders(transactions, prices)
portfolios[portfolio_name]['best'] = max(p_returns)
portfolios[portfolio_name]['worst'] = min(p_returns)
portfolios[portfolio_name]['trades'] = len(portfolios[portfolio_name]['orders'])
if path != None :
portfolios[portfolio_name].equity.to_csv(path + portfolio_name + '_equity.csv')
portfolios[portfolio_name].weight.to_csv(path + portfolio_name + '_weight.csv')
portfolios[portfolio_name].share.to_csv(path + portfolio_name + '_share.csv')
portfolios[portfolio_name].transactions.to_csv(path + portfolio_name + '_transactions.csv')
portfolios[portfolio_name].orders.to_csv(path + portfolio_name + '_orders.csv')
return
def Profit_Evaluation(act, ts, ts_pct_change, kind='s'):
'''
收益率评价
'''
act_ts = pd.Series(act[:-1], index=ts.index)
profit = []
buy = 0
sell = 0
flag = 0
buy_flag = 0
for i in range(len(act_ts)):
day = ts.index[i]
if flag == 0 and act_ts[i] == -1:
profit.append(0)
flag = 1
continue
elif act_ts[i] == 1 and ts_pct_change[day] < 9.99:
buy = ts[i]
profit.append(0)
flag = 1
buy_flag = 1
elif act_ts[i] == -1 and ts_pct_change[day] < 9.99:
sell = ts[i]
if kind == 's':
profit.append(0.9985 * (sell - buy) / buy)
elif kind == 'f':
profit.append(0.9995 * (sell - buy) / buy)
else:
profit.append((sell - buy) / buy)
buy_flag = 0
else:
profit.append(0)
cumprod_profit = np.cumproduct(np.array(profit) + 1)
profit_ts = pd.Series(profit, ts.index)
cumprod_profit_ts = pd.Series(cumprod_profit, ts.index)
buy_flag = 0
for i in range(len(cumprod_profit_ts)):
if act[i] == 1:
buy_flag = 1
if act[i] == -1:
buy_flag = 0
if buy_flag == 1:
cumprod_profit_ts[i] = cumprod_profit_ts[i - 1] * (ts[i] /
ts[i - 1])
return profit_ts, cumprod_profit_ts
def Tuple_MI(Tuple, IdxLength):
"""
Function to return the absolution position of a multiindex when the index tuple
and the index hierarchy and size are given.
Example: Tuple_MI([2,7,3],[100,10,5]) = 138
Tuple_MI is the inverse of MI_Tuple.
"""
# First, generate the index position offset values
A = IdxLength[1:] + IdxLength[:1] # Shift 1 to left
A[-1] = 1 # Replace lowest index by 1
A.reverse()
IdxPosOffset = np.cumproduct(A).tolist()
IdxPosOffset.reverse()
Position = np.sum([a * b for a, b in zip(Tuple, IdxPosOffset)])
return Position
def get_monthly_saving(age, retirement_age, retirement_saving_goal, ann_int,
annual_saving_growth):
"""
Given a constant of target cash at retirement, calculate a fix saving per month
"""
months_before_retirement = (retirement_age - age) * 12
# investment before retirement
ann_int = ann_int[:, :months_before_retirement]
interest_mlp_after_retirement_arr = np.cumproduct(1 + ann_int, axis=1)
## solve for constant saving per month
"""
retirement_saving_goal = sum(distcount_factors_array * saving_per_month_now)
retirement_saving_goal = sum(distcount_factors_array) * saving_per_month_now
"""
saving_per_month = retirement_saving_goal / np.sum(
interest_mlp_after_retirement_arr, axis=1)
# growth factor
num_sim = ann_int.shape[0]
saving_growth_multipliers = np.repeat(np.cumproduct(
np.ones(shape=(num_sim, np.int(months_before_retirement / 12))) *
(1 + annual_saving_growth),
axis=1),
12,
axis=1)
saving_first_month = retirement_saving_goal / np.sum(
interest_mlp_after_retirement_arr * saving_growth_multipliers, axis=1)
growth_saving = np.repeat(saving_first_month.reshape(
len(saving_first_month), 1),
months_before_retirement,
axis=1) * saving_growth_multipliers
return saving_per_month, growth_saving
def calculate_with_n_electron(self, phi, partition_function,
number_density, n_electron):
ion_populations = pd.DataFrame(data=0.0,
index=partition_function.index.copy(),
columns=partition_function.columns.copy(), dtype=np.float64)
for atomic_number, groups in phi.groupby(level='atomic_number'):
current_phis = (groups / n_electron).replace(np.nan, 0.0).values
phis_product = np.cumproduct(current_phis, axis=0)
neutral_atom_density = (number_density.ix[atomic_number] /
(1 + np.sum(phis_product, axis=0)))
ion_populations.ix[atomic_number, 0] = (
neutral_atom_density.values)
ion_populations.ix[atomic_number].values[1:] = (
neutral_atom_density.values * phis_product)
ion_populations[ion_populations < self.ion_zero_threshold] = 0.0
return ion_populations
def construct_zernike_lookuptable(zernike_indexes):
"""Return a lookup table of the sum-of-factorial part of the radial
polynomial of the zernike indexes passed
zernike_indexes - an Nx2 array of the Zernike polynomials to be
computed.
"""
factorial = np.ones((100,))
factorial[1:] = np.cumproduct(np.arange(1, 100).astype(float))
width = int(np.max(zernike_indexes[:,0]) / 2+1)
lut = np.zeros((zernike_indexes.shape[0],width))
for idx,(n,m) in zip(range(zernike_indexes.shape[0]),zernike_indexes):
for k in range(0,(n-m)/2+1):
lut[idx,k] = \
(((-1)**k) * factorial[n-k] /
(factorial[k]*factorial[(n+m)/2-k]*factorial[(n-m)/2-k]))
return lut
def backtest_old(prices, weights, period_ends, capital, offset=0., commission=0.) :
p_holdings = (capital / prices * weights.align(prices)[0]).shift(offset).ffill().fillna(0)
w = weights.align(prices)[0].shift(offset).fillna(0)
trade_dates = w[w.sum(1) != 0].index
p_cash = capital - (p_holdings * prices.shift(offset)).sum(1)
totalcash = p_cash[trade_dates].align(prices[prices.columns[0]])[0].ffill().fillna(0)
p_returns = (totalcash + (p_holdings * prices).sum(1) - \
(abs(p_holdings - p_holdings.shift(1)) * commission).sum(1)) / \
(totalcash + (p_holdings * prices.shift(1)).sum(1)) - 1
p_returns = p_returns.fillna(0)
# p_weights = p_holdings * prices.shift(offset) / (totalcash + (p_holdings * prices.shift(offset)).sum(1))
p_weights = pd.DataFrame([(p_holdings * prices.shift(offset))[symbol] / \
(totalcash + (p_holdings * prices.shift(offset)).sum(1)) \
for symbol in prices.columns], index=prices.columns).T
p_weights = p_weights.fillna(0)
return np.cumproduct(1. + p_returns) * capital, p_holdings.astype(int), p_returns, p_weights
def calc_exposed_faces(self):
#TODO: The 3D bitwise ops are slow
t = time.time()
air = BLOCK_SOLID[self.blocks] == 0
light = numpy.cumproduct(air[:,::-1,:], axis=1)[:,::-1,:]
exposed = numpy.zeros(air.shape,dtype=numpy.uint8)
exposed[0,:,:] |= self.edge_blocks(dx=-1)<<5 #left edge
exposed[-1,:,:] |= self.edge_blocks(dx=1)<<4 #right edge
exposed[:,:,0] |= self.edge_blocks(dz=-1)<<2 #back edge
exposed[:,:,-1] |= self.edge_blocks(dz=1)<<3 #front edge
exposed[:,:-1,:] |= air[:,1:,:]<<7 #up
exposed[:,1:,:] |= air[:,:-1,:]<<6 #down
exposed[1:,:,:] |= air[:-1,:,:]<<5 #left
exposed[:-1,:,:] |= air[1:,:,:]<<4 #right
exposed[:,:,:-1] |= air[:,:,1:]<<3 #forward
exposed[:,:,1:] |= air[:,:,:-1]<<2 #back
self.exposed = exposed*(~air)
def alpha(self, pH):
'''Return the fraction of each species at a given pH.
This returns a Numpy list of fractional speciation at a given
solution. The returned list will be ordered as per the Ka/pKa values,
with the most acidic component listed first.
'''
# If the given pH is not a list/array, be sure to convert it to one
# for future calcs.
if isinstance(pH, (int, float)):
pH = [pH,]
pH = np.array(pH, dtype=float)
# Calculate the concentration of H3O+. If multiple pH values are
# given, then it is best to construct a two dimensional array of
# concentrations.
h3o = 10.**(-pH)
if len(h3o) > 1:
h3o = np.repeat( h3o.reshape(-1, 1), len(self._Ka_temp), axis=1)
# These are the powers that the H3O+ concentrations will be raised.
power = np.arange(len(self._Ka_temp))
# Calculate the H3O+ concentrations raised to the powers calculated
# above (in reverse order).
h3o_pow = h3o**( power[::-1] )
# Calculate a cumulative product of the Ka values. The first value
# must be 1.0, which is why _Ka_temp is used instead of Ka.
Ka_prod = np.cumproduct(self._Ka_temp)
# Multiply the H3O**power values times the cumulative Ka product.
h3o_Ka = h3o_pow*Ka_prod
# Return the alpha values. The return signature will differ is the
# shape of the H3O array was 2-dimensional.
if len(h3o.shape) > 1:
den = h3o_Ka.sum(axis=1)
return h3o_Ka/den.reshape(-1,1)
else:
den = h3o_Ka.sum()
return h3o_Ka/den
def cartesian_product(X):
"""
Numpy version of itertools.product or pandas.compat.product.
Sometimes faster (for large inputs)...
Examples
--------
>>> cartesian_product([list('ABC'), [1, 2]])
[array(['A', 'A', 'B', 'B', 'C', 'C'], dtype='|S1'),
array([1, 2, 1, 2, 1, 2])]
"""
lenX = np.fromiter((len(x) for x in X), dtype=int)
cumprodX = np.cumproduct(lenX)
a = np.roll(cumprodX, 1)
a[0] = 1
b = cumprodX[-1] / cumprodX
return [np.tile(np.repeat(np.asarray(com._values_from_object(x)), b[i]), np.product(a[i])) for i, x in enumerate(X)]
def combinations(self, lol): ''' Take a list of lists of numbers. Return an array of float64, with shape NxM where N = cumprod([len(x) for x in lol]) and M = len(lol). The return array contains all possible combinations of the inputs (eg, sets of items containing one item from each of the lists in lol) ''' m=len(lol) shapes=[len(x) for x in lol] stride=np.cumproduct(shapes) n = stride[-1] oa=np.zeros((n, m), np.float64) oa[:,0]=np.resize(lol[0], n) for i in range(1, m): dl=stride[i-1] dr=n/dl oa[:,i]=np.ravel(np.transpose(np.resize(lol[i], (dl, dr)))) return oa
def index_data(self, coords, hyper_points, coord_map):
"""
Index the data that falls inside the grid cells
:param coords: coordinates of grid
:param hyper_points: list of HyperPoints to index
:param coord_map: list of tuples relating index in HyperPoint to index in coords and in
coords to be iterated over
"""
# create bounds in correct order
hp_coords = []
coord_descreasing = [False] * len(coords)
coord_lengths = [0] * len(coords)
lower_bounds = [None] * len(coords)
max_bounds = [None] * len(coords)
for (hpi, ci, shi) in coord_map:
coord = coords[ci]
# Coordinates must be monotonic; determine whether increasing or decreasing.
if len(coord.points) > 1:
if coord.points[1] < coord.points[0]:
coord_descreasing[shi] = True
coord_lengths[shi] = len(coord.points)
if coord_descreasing[shi]:
lower_bounds[shi] = coord.bounds[::-1, 1]
max_bounds[shi] = coord.bounds[0, 1]
else:
lower_bounds[shi] = coord.bounds[::, 0]
max_bounds[shi] = coord.bounds[-1, 1]
hp_coord = hyper_points.coords[hpi]
if isinstance(hp_coord[0], datetime.datetime):
hp_coord = convert_datetime_to_std_time(hp_coord)
hp_coords.append(hp_coord)
bounds_coords_max = list(zip(lower_bounds, hp_coords, max_bounds))
# stack for each coordinate
# where the coordinate is larger than the maximum set to -1
# otherwise search in the sorted coordinate to find all the index of the hyperpoints
# The choice of 'left' or 'right' and '<' and '<=' determines which
# cell is chosen when the coordinate is equal to the boundary.
# -1 or M_i indicates the point is outside the grid.
# Output is a list of coordinates which lists the indexes where the hyper points
# should be located in the grid
indices = np.vstack(
np.where(
ci < max_coordinate_value,
np.searchsorted(bi, ci, side='right') - 1,
-1)
for bi, ci, max_coordinate_value in bounds_coords_max)
# D-tuple giving the shape of the output grid
grid_shape = tuple(len(bi_ci[0]) for bi_ci in bounds_coords_max)
# shape (N,) telling which points actually fall within the grid,
# i.e. have indexes that are not -1 and are not masked data points
grid_mask = np.all(
(indices >= 0) &
(ma.getmaskarray(hyper_points.data) == False),
axis=0)
# if the coordinate was decreasing then correct the indices for this cell
for indices_slice, decreasing, coord_length in zip(range(indices.shape[0]), coord_descreasing, coord_lengths):
if decreasing:
# indices[indices_slice] += (coord_length - 1) - indices[indices_slice]
indices[indices_slice] *= -1
indices[indices_slice] += (coord_length - 1)
# shape (N,) containing negative scalar cell numbers for each
# input point (sequence doesn't matter so long as they are unique), or
# -1 for points outside the grid.
#
# Possibly numpy.lexsort could be used to avoid the need for this,
# although we'd have to be careful about points outside the grid.
self.cell_numbers = np.where(
grid_mask,
np.tensordot(
np.cumproduct((1,) + grid_shape[:-1]),
indices,
axes=1
),
-1)
# Sort everything by cell number
self.sort_order = np.argsort(self.cell_numbers)
self.cell_numbers = self.cell_numbers[self.sort_order]
self._indices = indices[:, self.sort_order]
self.hp_coords = [hp_coord[self.sort_order] for hp_coord in hp_coords]
def cumproduct(x, axis=0):
return np.cumproduct(x, axis)
