def _fit_eig(self, x):
scatter_matrix = _scatter_matrix(x, self.arity)
cov_matrix = _estimate_covariance(scatter_matrix, x.shape[0])
if self.n_components:
shape1 = self.n_components
else:
shape1 = x.shape[1]
n_blocks = int(ceil(shape1 / x._reg_shape[1]))
val_blocks = Array._get_out_blocks((1, n_blocks))
vec_blocks = Array._get_out_blocks((n_blocks, x._n_blocks[1]))
_decompose(cov_matrix, self.n_components, x._reg_shape[1],
val_blocks,
vec_blocks)
bshape = (x._reg_shape[1], x._reg_shape[1])
self.components_ = Array(vec_blocks, bshape, bshape,
(shape1, x.shape[1]), False)
self.explained_variance_ = Array(val_blocks, bshape, bshape,
(1, shape1), False)
return self
def _sort_v(v, sorting):
v_blocks = [[] for _ in range(v._n_blocks[1])]
hbsize = v._reg_shape[1]
for i, vblock in enumerate(v._iterator("columns")):
out_blocks = [[] for _ in range(v._n_blocks[1])]
_sort_v_block(vblock._blocks, i, hbsize, sorting, out_blocks)
for j in range(len(out_blocks)):
v_blocks[j].append(out_blocks[j])
vbsize = v._reg_shape[0]
final_blocks = Array._get_out_blocks(v._n_blocks)
for i, v_block in enumerate(v_blocks):
new_block = [object() for _ in range(v._n_blocks[0])]
_merge_svd_block(v_block, i, hbsize, vbsize, sorting, new_block)
for j in range(len(new_block)):
final_blocks[j][i] = new_block[j]
for elem in v_block:
compss_delete_object(elem)
return Array(final_blocks, v._top_left_shape, v._reg_shape, v.shape,
v._sparse)
def _compute_u_sorted(a, sorting):
u_blocks = [[] for _ in range(a._n_blocks[1])]
hbsize = a._reg_shape[1]
for i, vblock in enumerate(a._iterator("columns")):
u_block = [object() for _ in range(a._n_blocks[1])]
_compute_u_block_sorted(vblock._blocks, i, hbsize, sorting, u_block)
for j in range(len(u_block)):
u_blocks[j].append(u_block[j])
vbsize = a._reg_shape[0]
final_blocks = Array._get_out_blocks(a._n_blocks)
for i, u_block in enumerate(u_blocks):
new_block = [object() for _ in range(a._n_blocks[0])]
_merge_svd_block(u_block, i, hbsize, vbsize, sorting, new_block)
for j in range(len(new_block)):
final_blocks[j][i] = new_block[j]
for elem in u_block:
compss_delete_object(elem)
return Array(final_blocks, a._top_left_shape, a._reg_shape, a.shape,
a._sparse)
def _update_u(z, u_blocks, w_blocks, out_blocks):
u_np = np.squeeze(Array._merge_blocks(u_blocks))
w_np = np.squeeze(Array._merge_blocks(w_blocks))
u_new = u_np + w_np - z
n_cols = u_blocks[0][0].shape[1]
for i in range(len(out_blocks)):
out_blocks[i] = u_new[i * n_cols:(i + 1) * n_cols].reshape(1, -1)
def _partial_variability_params(x, y, mean_x, mean_y):
x, y = Array._merge_blocks(x), Array._merge_blocks(y)
normalized_x = x[:, 0] - mean_x # the 0 is because only 1D LR is supported
normalized_y = y - mean_y
normalized_xy_dot = np.dot(normalized_x, normalized_y)
normalized_xx_dot = np.dot(normalized_x, normalized_x)
return normalized_xy_dot, normalized_xx_dot
def _score(x_list, y_list, clf):
x = Array._merge_blocks(x_list)
y = Array._merge_blocks(y_list)
y_pred = clf.predict(x)
equal = np.equal(y_pred, y.ravel())
return np.sum(equal), x.shape[0]
def kneighbors(self, x, n_neighbors=None, return_distance=True):
""" Finds the K nearest neighbors of the input samples. Returns
indices and distances to the neighbors of each sample.
Parameters
----------
x : ds-array, shape=(n_samples, n_features)
The query samples.
n_neighbors: int, optional (default=None)
Number of neighbors to get. If None, the value passed in the
constructor is employed.
return_distance : boolean, optional (default=True)
Whether to return distances.
Returns
-------
dist : ds-array, shape=(n_samples, n_neighbors)
Array representing the lengths to points, only present if
return_distance=True.
ind : ds-array, shape=(n_samples, n_neighbors)
Indices of the nearest samples in the fitted data.
"""
validation.check_is_fitted(self, '_fit_data')
if n_neighbors is None:
n_neighbors = self.n_neighbors
distances = []
indices = []
for q_row in x._iterator(axis=0):
queries = []
for row in self._fit_data._iterator(axis=0):
queries.append(
_get_neighbors(row._blocks, q_row._blocks, n_neighbors))
dist, ind = _merge_queries(*queries)
distances.append([dist])
indices.append([ind])
ind_arr = Array(blocks=indices,
top_left_shape=(x._top_left_shape[0], n_neighbors),
reg_shape=(x._reg_shape[0], n_neighbors),
shape=(x.shape[0], n_neighbors),
sparse=False)
if return_distance:
dst_arr = Array(blocks=distances,
top_left_shape=(x._top_left_shape[0], n_neighbors),
reg_shape=(x._reg_shape[0], n_neighbors),
shape=(x.shape[0], n_neighbors),
sparse=False)
return dst_arr, ind_arr
return ind_arr
def _partial_estimate_parameters(x, resp):
x = Array._merge_blocks(x)
resp = Array._merge_blocks(resp)
partial_nk = resp.sum(axis=0)
if issparse(x):
partial_means = x.T.dot(resp).T
else:
partial_means = np.matmul(resp.T, x)
return x.shape[0], partial_nk, partial_means
def _partial_covar_diag(resp, x, means):
x = Array._merge_blocks(x)
resp = Array._merge_blocks(resp)
if issparse(x):
avg_resp_sample_2 = x.multiply(x).T.dot(resp).T
avg_sample_means = means * x.T.dot(resp).T
else:
avg_resp_sample_2 = np.dot(resp.T, x * x)
avg_sample_means = means * np.dot(resp.T, x)
return avg_resp_sample_2 - 2 * avg_sample_means
def _choose_and_assign_rows_xy(x, y, subsamples_sizes, subsamples, seed):
np.random.seed(seed)
x = Array._merge_blocks(x)
y = Array._merge_blocks(y)
indices = np.random.permutation(x.shape[0])
start = 0
for i, size in enumerate(subsamples_sizes):
end = start + size
subsamples[i] = (x[indices[start:end]], y[indices[start:end]])
start = end
def _generate_bins(mn, mx, dimensions, n_regions):
bins = []
mn_arr = Array._merge_blocks(mn)[0]
mx_arr = Array._merge_blocks(mx)[0]
# create bins for the different regions in the grid in every dimension
for dim in dimensions:
bin_ = np.linspace(mn_arr[dim], mx_arr[dim], n_regions + 1)
bins.append(bin_)
return bins
def _get_neighbors(blocks, q_blocks, n_neighbors):
samples = Array._merge_blocks(blocks)
q_samples = Array._merge_blocks(q_blocks)
n_samples = samples.shape[0]
knn = SKNeighbors(n_neighbors=n_neighbors)
knn.fit(X=samples)
dist, ind = knn.kneighbors(X=q_samples)
return dist, ind, n_samples
def _u_step(self):
u_blocks = []
for u_hblock, w_hblock in zip(self._u._iterator(),
self._w._iterator()):
out_blocks = [object() for _ in range(self._u._n_blocks[1])]
_update_u(self._z, u_hblock._blocks, w_hblock._blocks, out_blocks)
u_blocks.append(out_blocks)
r_shape = self._u._reg_shape
shape = self._u.shape
self._u = Array(u_blocks, r_shape, r_shape, shape, self._u._sparse)
def _transform(blocks, m_blocks, v_blocks, out_blocks):
x = Array._merge_blocks(blocks)
mean = Array._merge_blocks(m_blocks)
var = Array._merge_blocks(v_blocks)
scaled_x = (x - mean) / np.sqrt(var)
constructor_func = np.array if not issparse(x) else csr_matrix
start, end = 0, 0
for i, block in enumerate(blocks[0]):
end += block.shape[1]
out_blocks[i] = constructor_func(scaled_x[:, start:end])
def _merge(x_list, y_list, id_list):
samples = Array._merge_blocks(x_list)
labels = Array._merge_blocks(y_list)
sample_ids = Array._merge_blocks(id_list)
_, uniques = np.unique(sample_ids, return_index=True)
indices = np.argsort(uniques)
uniques = uniques[indices]
sample_ids = sample_ids[uniques]
samples = samples[uniques]
labels = labels[uniques]
return samples, labels, sample_ids
def _compute_primal_res(self, z_old):
blocks = []
for w_hblock in self._w._iterator():
out_blocks = [object() for _ in range(self._w._n_blocks[1])]
_substract(w_hblock._blocks, z_old, out_blocks)
blocks.append(out_blocks)
prires = Array(blocks, self._w._reg_shape, self._w._reg_shape,
self._w.shape, self._w._sparse)
# this should be a ds-array of a single element. We return only the
# block
return (prires.norm(axis=1)**2).sum().sqrt()
def transform_to_rf_dataset(x: Array, y: Array) -> RfDataset:
"""Creates a RfDataset object from samples x and labels y.
This function creates a dislib.classification.rf.data.RfDataset by saving
x and y in files.
Parameters
----------
x : ds-array, shape = (n_samples, n_features)
The training input samples.
y : ds-array, shape = (n_samples,) or (n_samples, n_outputs)
The target values.
Returns
-------
rf_dataset : dislib.classification.rf._data.RfDataset
"""
n_samples = x.shape[0]
n_features = x.shape[1]
samples_file = tempfile.NamedTemporaryFile(mode='wb',
prefix='tmp_rf_samples_',
delete=False)
samples_path = samples_file.name
samples_file.close()
_allocate_samples_file(samples_path, n_samples, n_features)
start_idx = 0
row_blocks_iterator = x._iterator(axis=0)
top_row = next(row_blocks_iterator)
_fill_samples_file(samples_path, top_row._blocks, start_idx)
start_idx += x._top_left_shape[0]
for x_row in row_blocks_iterator:
_fill_samples_file(samples_path, x_row._blocks, start_idx)
start_idx += x._reg_shape[0]
labels_file = tempfile.NamedTemporaryFile(mode='w',
prefix='tmp_rf_labels_',
delete=False)
labels_path = labels_file.name
labels_file.close()
for y_row in y._iterator(axis=0):
_fill_labels_file(labels_path, y_row._blocks)
rf_dataset = RfDataset(samples_path, labels_path)
rf_dataset.n_samples = n_samples
rf_dataset.n_features = n_features
return rf_dataset
def _compute_var(blocks, m_blocks):
x = Array._merge_blocks(blocks)
mean = Array._merge_blocks(m_blocks)
sparse = issparse(x)
if sparse:
x = x.toarray()
mean = mean.toarray()
var = np.mean(np.array(x - mean)**2, axis=0)
if sparse:
return csr_matrix(var)
else:
return var
def _soft_thresholding(w_blocks, u_blocks, k):
w_mean = np.squeeze(Array._merge_blocks(w_blocks))
u_mean = np.squeeze(Array._merge_blocks(u_blocks))
v = w_mean + u_mean
z = np.zeros(v.shape)
for i in range(z.shape[0]):
if np.abs(v[i]) <= k:
z[i] = 0
else:
if v[i] > k:
z[i] = v[i] - k
else:
z[i] = v[i] + k
return z
def _partial_covar_full(resp, x, means):
x = Array._merge_blocks(x)
resp = Array._merge_blocks(resp)
n_components, n_features = means.shape
covariances = np.empty((n_components, n_features, n_features))
for k in range(n_components):
if issparse(x):
diff = (x - means[k] for x in x)
partial_covs = (np.dot(r * d.T, d) for d, r in
zip(diff, resp[:, k]))
covariances[k] = sum(partial_covs)
else:
diff = x - means[k]
covariances[k] = np.dot(resp[:, k] * diff.T, diff)
return covariances
def transform(self, x):
"""
Standarize data.
Parameters
----------
x : ds-array, shape=(n_samples, n_features)
Returns
-------
x_new : ds-array, shape=(n_samples, n_features)
Scaled data.
"""
if self.mean_ is None or self.var_ is None:
raise Exception("Model has not been initialized.")
n_blocks = x._n_blocks[1]
blocks = []
m_blocks = self.mean_._blocks
v_blocks = self.var_._blocks
for row in x._iterator(axis=0):
out_blocks = [object() for _ in range(n_blocks)]
_transform(row._blocks, m_blocks, v_blocks, out_blocks)
blocks.append(out_blocks)
return Array(blocks,
top_left_shape=x._top_left_shape,
reg_shape=x._reg_shape,
shape=x.shape,
sparse=x._sparse)
def fit(self, x, y):
"""
Fits the model with training data.
Parameters
----------
x : ds-array, shape=(n_samples, n_features)
Training samples.
y : ds-array, shape=(n_samples, 1)
Class labels of x.
Returns
-------
self : ADMM
"""
if not x._is_regular():
x_reg = x.rechunk(x._reg_shape)
else:
x_reg = x
self._init_model(x_reg)
while not self.converged_ and self.n_iter_ < self.max_iter:
self._step(x_reg, y)
self.n_iter_ += 1
if self.verbose:
print("Iteration ", self.n_iter_)
z_blocks = [object() for _ in range(x_reg._n_blocks[1])]
_split_z(self._z, x._reg_shape[1], z_blocks)
self.z_ = Array([z_blocks], (1, x._reg_shape[1]), (1, x._reg_shape[1]),
(1, x.shape[1]), False)
return self
def _partial_covar_tied(x):
x = Array._merge_blocks(x)
if issparse(x):
avg_sample_2 = x.T.dot(x)
else:
avg_sample_2 = np.dot(x.T, x)
return avg_sample_2
def predict(self, x):
"""
Predict using the linear model.
Parameters
----------
x : ds-array
Samples to be predicted: x.shape (n_samples, 1).
Returns
-------
y : ds-array
Predicted values
"""
blocks = [list()]
for r_block in x._iterator(axis='rows'):
blocks[0].append(
_predict(r_block._blocks, self.coef_, self.intercept_))
return Array(blocks=blocks,
top_left_shape=(x._top_left_shape[0], 1),
reg_shape=(x._reg_shape[0], 1),
shape=(x.shape[0], 1),
sparse=x._sparse)
def _rotate(coli_blocks, colj_blocks, j):
if j is None:
return
coli = Array._merge_blocks(coli_blocks)
colj = Array._merge_blocks(colj_blocks)
n = coli.shape[1]
coli_k = coli @ j[:n, :n] + colj @ j[n:, :n]
colj_k = coli @ j[:n, n:] + colj @ j[n:, n:]
block_size = coli_blocks[0][0].shape[0]
for i in range(len(coli_blocks)):
coli_blocks[i][0][:] = coli_k[i * block_size:(i + 1) * block_size][:]
colj_blocks[i][0][:] = colj_k[i * block_size:(i + 1) * block_size][:]
def _subset_transform(blocks, u_blocks, c_blocks, reg_shape, out_blocks):
data = Array._merge_blocks(blocks)
mean = Array._merge_blocks(u_blocks)
components = Array._merge_blocks(c_blocks)
if issparse(data):
data = data.toarray()
mean = mean.toarray()
res = (np.matmul(data - mean, components.T))
if issparse(data):
res = csr_matrix(res)
for j in range(0, len(blocks[0])):
out_blocks[j] = res[:, j * reg_shape:(j + 1) * reg_shape]
def _subset_scatter_matrix(blocks):
data = Array._merge_blocks(blocks)
if issparse(data):
data = data.toarray()
return np.dot(data.T, data)
def predict(self, x):
""" Perform classification on samples.
Parameters
----------
x : ds-array, shape=(n_samples, n_features)
Input samples.
Returns
-------
y : ds-array, shape(n_samples, 1)
Class labels of x.
"""
assert (self._clf is not None or self._svs is not None), \
"Model has not been initialized. Call fit() first."
y_list = []
for row in x._iterator(axis=0):
y_list.append([_predict(row._blocks, self._clf)])
return Array(blocks=y_list,
top_left_shape=(x._top_left_shape[0], 1),
reg_shape=(x._reg_shape[0], 1),
shape=(x.shape[0], 1),
sparse=False)
def _load_mdcrd(path, block_size, n_cols, n_blocks, bytes_per_snap,
bytes_per_block):
blocks = []
file_size = os.stat(path).st_size - _CRD_LINE_SIZE
try:
fid = open(path, "rb")
fid.read(_CRD_LINE_SIZE) # skip header
for _ in range(0, file_size, bytes_per_block):
data = fid.read(bytes_per_block)
out_blocks = [object() for _ in range(n_blocks)]
_read_crd_bytes(data, block_size[1], n_cols, out_blocks)
compss_delete_object(data)
blocks.append(out_blocks)
finally:
fid.close()
n_samples = int(file_size / bytes_per_snap)
return Array(blocks,
top_left_shape=block_size,
reg_shape=block_size,
shape=(n_samples, n_cols),
sparse=False)
def decision_function(self, x):
""" Evaluates the decision function for the samples in x.
Parameters
----------
x : ds-array, shape=(n_samples, n_features)
Input samples.
Returns
-------
df : ds-array, shape=(n_samples, 2)
The decision function of the samples for each class in the model.
"""
assert (self._clf is not None or self._svs is not None), \
"Model has not been initialized. Call fit() first."
df = []
for row in x._iterator(axis=0):
df.append([_decision_function(row._blocks, self._clf)])
return Array(blocks=df,
top_left_shape=(x._top_left_shape[0], 1),
reg_shape=(x._reg_shape[0], 1),
shape=(x.shape[0], 1),
sparse=False)
