def evaluate_item_with_code(train: ss.csr_matrix,
test: ss.csr_matrix,
user: np.ndarray,
item_code: np.ndarray,
item_center: List[np.ndarray],
topk=200,
cutoff=200):
train = train.tocsr()
test = test.tocsr()
#result1 = Eval.topk_search_with_code(train, user, item_code, item_center, topk)
result = Eval.topk_search_with_code_fast(train, user, item_code,
item_center, topk)
return Eval.evaluate_topk(train, test, result, cutoff)
def evaluate_item(train:ss.csr_matrix, test:ss.csr_matrix, user:np.ndarray, item:np.ndarray, topk:int=200, cutoff:int=200):
train = train.tocsr()
test = test.tocsr()
idx = np.squeeze((test.sum(axis=1) > 0).A)
train = train[idx, :]
test = test[idx, :]
user = user[idx, :]
N = train.shape[1]
cand_count = N - train.sum(axis=1)
if topk <0:
mat_rank = Eval.predict(train, test, user, item)
else:
mat_rank = Eval.topk_search_(train, test, user, item, topk)
return Eval.compute_item_metric(test, mat_rank, cand_count, cutoff)
def evaluate_item(train: ss.csr_matrix,
test: ss.csr_matrix,
user: np.ndarray,
item: np.ndarray,
topk: int = -1,
cutoff: int = 100):
train = train.tocsr()
test = test.tocsr()
N = train.shape[1]
cand_count = N - train.sum(axis=1)
if topk < 0:
mat_rank = Eval.predict(train, test, user, item)
else:
mat_rank = Eval.topk_search_(train, test, user, item, topk)
return Eval.compute_item_metric(test, mat_rank, cand_count, cutoff)
def _save(
filepath: Union[str, Path],
matrix: csr_matrix,
cell_ids: pd.DataFrame,
features: pd.DataFrame,
save_pickle: bool = False,
save_rds: bool = False,
save_h5ad: bool = False,
save_loom: bool = False,
meta: Optional[pd.DataFrame] = None,
):
filepath = Path(filepath)
if save_pickle:
build_counts_store(matrix.tocoo(),
cell_ids,
features,
save_path=filepath)
if save_rds:
Convert.pickle_to_rds_dir(filepath.parent)
if save_h5ad or save_loom:
cell_ids = meta if meta is not None else pd.DataFrame(
cell_ids).set_index(0)
cell_ids.index.name = "index"
features = features.rename(columns={
"ensgs": "gene_ids",
"genes": "index"
}).set_index("index")
adata = AnnData(matrix.tocsr(), cell_ids, features)
if save_h5ad:
adata.write_h5ad(filepath.parent / "rna.h5ad")
if save_loom:
adata.write_loom(filepath.parent / "rna.loom")
def evaluate_topk(train:ss.csr_matrix, test:ss.csr_matrix, topk_item:np.ndarray, cutoff:int=200):
train = train.tocsr()
test = test.tocsr()
result = topk_item
N = train.shape[1]
cand_count = N - train.sum(axis=1)
M = test.shape[0]
uir = []
for i in range(M):
R = set(test.indices[test.indptr[i]:test.indptr[i+1]])
for k in range(result.shape[1]):
if result[i,k] in R:
uir.append((i, result[i,k], k))
user_id, item_id, rank = zip(*uir)
mat_rank = ss.csr_matrix((rank, (user_id, item_id)), shape=test.shape)
return Eval.compute_item_metric(test, mat_rank, cand_count, cutoff)
def neighbor_sampler(adj_matrix: sp.csr_matrix, max_degree: int = 25,
selfloop: bool = False):
adj_matrix = adj_matrix.tocsr(copy=False)
N = adj_matrix.shape[0]
neighbors_matrix = N * np.ones((N + 1, max_degree), dtype=intx())
for nodeid in range(N):
neighbors = adj_matrix[nodeid].indices
# if not selfloop:
# neighbors = np.setdiff1d(neighbors, [nodeid])
# else:
# neighbors = np.intersect1d(neighbors, [nodeid])
size = neighbors.size
if size == 0:
continue
if size > max_degree:
neighbors = np.random.choice(neighbors, max_degree, replace=False)
elif size < max_degree:
neighbors = np.random.choice(neighbors, max_degree, replace=True)
neighbors_matrix[nodeid] = neighbors
np.random.shuffle(neighbors_matrix.T)
return neighbors_matrix
def _k_neighbors_precomputed_sparse(self,
X: csr_matrix,
n_samples: int = None):
""" Find nearest neighbors in sparse distance matrix.
Parameters
----------
X: sparse, shape = [n_test, n_indexed]
Sparse distance matrix. Only non-zero elements
may be considered neighbors.
n_samples: int
Number of sampled indexed objects, e.g.
in approximate hubness reduction.
If None, this is inferred from the first row of X.
Returns
-------
k_neighbors : ndarray
Flattened array of neighbor indices.
"""
if not issparse(X):
raise TypeError(f'Matrix X is not sparse')
X = X.tocsr()
if n_samples is None:
n_samples = X.indptr[1] - X.indptr[0]
n_test, _ = X.shape
# To allow different number of explicit entries per row,
# we need to process the matrix row-by-row.
if np.all(X.indptr[1:] -
X.indptr[:-1] == n_samples) and not self.shuffle_equal:
min_ind = np.argpartition(X.data.reshape(n_test, n_samples),
kth=np.arange(self.k),
axis=1)[:, :self.k]
k_neighbors = X.indices[min_ind.ravel() +
np.repeat(X.indptr[:-1], repeats=self.k)]
else:
k_neighbors = np.empty((n_test, ), dtype=object)
if self.verbose:
range_n_test = tqdm(range(n_test))
else:
range_n_test = range(n_test)
if self.shuffle_equal:
for i in range_n_test:
x = X.getrow(i)
rp = self._random_state.permutation(x.nnz)
d2 = x.data[rp]
d2idx = np.argpartition(d2, kth=np.arange(self.k))
k_neighbors[i] = x.indices[rp[d2idx[:self.k]]]
else:
for i in range_n_test:
x = X.getrow(i)
min_ind = np.argpartition(x.data,
kth=np.arange(self.k))[:self.k]
k_neighbors[i] = x.indices[min_ind]
k_neighbors = np.concatenate(k_neighbors)
return k_neighbors
def compute_reg_scale(X: sps.csr_matrix, alpha0: float, nu: float) -> float:
X_csr: sps.csr_matrix = X.tocsr()
X_csr.sort_indices()
X_csc = X_csr.tocsc()
X_csc.sort_indices()
U, I = X.shape
nnz_row: np.ndarray = X_csr.indptr[1:] - X_csr.indptr[:-1]
nnz_col: np.ndarray = X_csc.indptr[1:] - X_csc.indptr[:-1]
return float(((nnz_row + alpha0 * I)**nu).sum()) + float(
((nnz_col + alpha0 * U)**nu).sum())
def topk_search(train:ss.csr_matrix, user:np.ndarray, item:np.ndarray, topk:int=200)->np.ndarray:
train = train.tocsr()
M, _ = train.shape
item_t = item.T
result = np.zeros((M, topk), dtype=np.int)
for i in range(M):
E = train.indices[train.indptr[i]:train.indptr[i+1]]
pred = np.matmul(user[i,:], item_t)
#pred = np.tensordot(user[i,:], item, [0,-1])
pred[E] = -np.inf
idx = np.argpartition(pred, -topk)[-topk:]
result[i,:] = idx[np.argsort(-pred[idx])]
return result
def _store_sparse(self, group: tables.Group, name: str,
arr: csr_matrix) -> None:
if not sparse.isspmatrix_csr(arr):
arr = arr.tocsr()
csr_group = self.h5f.create_group(group, name)
csr_group.was_sparse = True
if arr is not None and arr.nnz > 0:
self.h5f.create_array(csr_group, 'data', arr.data)
self.h5f.create_array(csr_group, 'indptr', arr.indptr)
self.h5f.create_array(csr_group, 'indices', arr.indices)
self.h5f.create_array(csr_group, 'shape', arr.shape)
self.h5f.flush()
def normalize_adj(adj : sp.csr_matrix):
"""Normalize adjacency matrix and convert it to a sparse tensor."""
if sp.isspmatrix(adj):
adj = adj.tolil()
adj.setdiag(1)
adj = adj.tocsr()
deg = np.ravel(adj.sum(1))
deg_sqrt_inv = 1 / np.sqrt(deg)
adj_norm = adj.multiply(deg_sqrt_inv[:, None]).multiply(deg_sqrt_inv[None, :])
elif torch.is_tensor(adj):
deg = adj.sum(1)
deg_sqrt_inv = 1 / torch.sqrt(deg)
adj_norm = adj * deg_sqrt_inv[:, None] * deg_sqrt_inv[None, :]
return to_sparse_tensor(adj_norm)
def _weight_matrix(dataMatrix: sps.csr_matrix,
weights: np.ndarray,
strategy="linear"):
"""
Assuming that rows are the object to weights, it returns a weighted matrix based on the weights and alpha
:param dataMatrix: dataMatrix with rows being the object to weight
:param weights: an array with the same length as the number of nnz inside the dataMatrix
:param strategy: strategy to use in order to put weights
:return: new data matrix with weighted data
"""
if len(weights) != len(dataMatrix.data):
raise ValueError(
"Demographic list does not contain all users in dataMatrix")
matrix = dataMatrix.tocsr(copy=True)
strategy_funct = STRATEGY_MAPPER[strategy]
matrix = strategy_funct.weight_matrix(matrix, feature_data=weights)
return sps.csr_matrix(matrix)
def _preprocess_URM_all(self, URM_all: sps.csr_matrix):
warm_items_mask = np.ediff1d(URM_all.tocsc().indptr) > self.threshold_items
self.warm_items = np.arange(URM_all.shape[1])[warm_items_mask]
URM_all = URM_all[:, self.warm_items]
warm_users_mask = np.ediff1d(URM_all.tocsr().indptr) > self.threshold_users
self.warm_users = np.arange(URM_all.shape[0])[warm_users_mask]
URM_all = URM_all[self.warm_users, :]
self._LOADED_GLOBAL_MAPPER_DICT["user_original_ID_to_index"] = reconcile_mapper_with_removed_tokens(
self._LOADED_GLOBAL_MAPPER_DICT["user_original_ID_to_index"],
np.arange(0, len(warm_users_mask), dtype=np.int)[np.logical_not(warm_users_mask)])
self._LOADED_GLOBAL_MAPPER_DICT["item_original_ID_to_index"] = reconcile_mapper_with_removed_tokens(
self._LOADED_GLOBAL_MAPPER_DICT["item_original_ID_to_index"],
np.arange(0, len(warm_items_mask), dtype=np.int)[np.logical_not(warm_items_mask)])
return URM_all
def with_cliques(adjacency: sp.csr_matrix,
clique_size: int,
num_cliques: int = 1) -> Tuple[sp.csr_matrix, np.ndarray]:
"""
Get adjacency matrix of dataset with cliques.
A clique is defined as a set of nodes where each node is a neighbor of every other
node.
Args:
adjacency: adjacency matrix to start with.
clique_size: size of each clique.
num_cliques: number of cliques to add.
Returns:
augmented_adjacency: adjacency with cliques added.
cliques: [num_cliques, clique_size] int32 array of indices of clique nodes.
"""
num_nodes = adjacency.shape[0]
adjacency = adjacency.tolil()
dtype = adjacency.dtype
rows = adjacency.rows
data = adjacency.data
cliques = np.empty((num_cliques, clique_size), dtype=np.int32)
for i in range(num_cliques):
clique = np.random.choice(num_nodes, clique_size, replace=False)
clique.sort()
cliques[i] = clique
for c in clique:
row = set(rows[c])
contains_c = c in row
row.update(clique)
if not contains_c:
row.remove(c)
rows[c] = sorted(row)
data[c] = np.ones((len(row), ), dtype=dtype)
return adjacency.tocsr(), cliques
def split(urm: sps.csr_matrix, range_intervals_list: list, k_out_list: list,
probability_list: list) -> (sps.csr_matrix, sps.csr_matrix):
# Initialize timer
start_time = time.time()
# Checking the size of the list
assert len(range_intervals_list) == len(k_out_list) == len(probability_list), \
"The input lists must have the same length"
# Generating a copy of the matrix
matrix = urm.tocsr(copy=True)
# Initializing the matrices
train_matrix = sps.lil_matrix(matrix.shape, dtype=np.float)
test_matrix = sps.lil_matrix(matrix.shape, dtype=np.float)
# Users
users = range(matrix.shape[0])
# Creating a list of triple (range_interval, k_out, probability)
triple_list = [
(r, k, p)
for r, k, p in zip(range_intervals_list, k_out_list, probability_list)
]
# Start splitting the input matrix
for user in users:
# Finding the items with which the user has interacted
items = matrix[user].indices
n_items = len(items)
# Finding the splitting category for the user
k, p = -1, -1
for range_interval, k_out, probability in triple_list:
if range_interval[0] <= n_items <= range_interval[1]:
k = k_out
p = probability
break
# Extracting k unique items (if possible)
n_items_to_be_extracted = min(k, n_items)
rnd.shuffle(items)
k_extracted_items = items[:n_items_to_be_extracted]
remaining_items = items[n_items_to_be_extracted:]
# Extracting elements with probability p
p_extracted_items = [
item for item in remaining_items if rnd.random() <= p
]
remaining_items = [
item for item in remaining_items if item not in p_extracted_items
]
# Incrementally generate fill the train and test matrix
for item in remaining_items:
train_matrix[user, item] = matrix[user, item]
for item in k_extracted_items:
test_matrix[user, item] = matrix[user, item]
for item in p_extracted_items:
test_matrix[user, item] = matrix[user, item]
# Converting train matrix and test matrix to csr format
train_matrix = train_matrix.tocsr()
test_matrix = test_matrix.tocsr()
# Checking output matrices
assert train_matrix.shape == test_matrix.shape == matrix.shape, \
"Train and test matrices should have the same shape as the input one"
assert (train_matrix.nnz + test_matrix.nnz) == matrix.nnz, \
f"The sum of train and test matrices nnz elements should be exactly the nnz elements in the input matrix:\n" \
f"train matrix nnz elements : {train_matrix.nnz}\n" \
f"test matrix nnz elements : {test_matrix.nnz}\n" \
f"matrix nnz elements : {matrix.nnz}\n"
# Ending time
end_time = time.time()
print(
f"Time taken to split the input matrix : {end_time - start_time} seconds"
)
return train_matrix, test_matrix
def __init__(self,
adj_matrix: sp.csr_matrix,
attr_matrix=None,
labels: np.ndarray = None,
node_names: np.ndarray = None,
attr_names: np.ndarray = None,
class_names: np.ndarray = None,
metadata: object = None):
"""Create attrribute graph
Parameters
----------
adj_matrix : sp.csr_matrix, shape [num_nodes, num_nodes]
Adjacency matrix in CSR format.
attr_matrix : sp.csr_matrix or np.ndarray, shape [num_nodes, num_attr], optional
Attribute matrix in CSR or numpy format.
labels : np.ndarray, shape [num_nodes], optional
Array, where each entry represents respective node's label(s).
node_names : np.ndarray, shape [num_nodes], optional
Names of nodes (as strings).
attr_names : np.ndarray, shape [num_attr]
Names of the attributes (as strings).
class_names : np.ndarray, shape [num_classes], optional
Names of the class labels (as strings).
metadata : object
Additional metadata such as text.
"""
if sp.isspmatrix(adj_matrix):
adj_matrix = adj_matrix.tocsr().astype(np.float32)
else:
raise ValueError(
"Adjacency matrix must be in sparse format (got {0} instead)".
format(type(adj_matrix)))
if adj_matrix.shape[0] != adj_matrix.shape[1]:
raise ValueError("Dimensions of the adjacency matrix don't agree")
if attr_matrix is not None:
if sp.isspmatrix(attr_matrix):
attr_matrix = attr_matrix.tocsr().astype(np.float32)
elif isinstance(attr_matrix, np.ndarray):
attr_matrix = attr_matrix.astype(np.float32)
else:
raise ValueError(
"Attribute matrix must be a sp.spmatrix or a np.ndarray (got {0} instead)"
.format(type(attr_matrix)))
if attr_matrix.shape[0] != adj_matrix.shape[0]:
raise ValueError(
"Dimensions of the adjacency and attribute matrices don't agree"
)
if labels is not None:
if labels.shape[0] != adj_matrix.shape[0]:
raise ValueError(
"Dimensions of the adjacency matrix and the label vector don't agree"
)
if node_names is not None:
if len(node_names) != adj_matrix.shape[0]:
raise ValueError(
"Dimensions of the adjacency matrix and the node names don't agree"
)
if attr_names is not None:
if len(attr_names) != attr_matrix.shape[1]:
raise ValueError(
"Dimensions of the attribute matrix and the attribute names don't agree"
)
self.adj_matrix = adj_matrix
self.attr_matrix = attr_matrix
self.labels = labels
self.node_names = node_names
self.attr_names = attr_names
self.class_names = class_names
self.metadata = metadata
def removeEye(adj: sp.csr_matrix):
adj = adj.tolil(copy=True)
adj.setdiag(0)
return adj.tocsr()
def addEye(adj: sp.csr_matrix):
adj = adj.tolil(copy=True)
adj.setdiag(1)
return adj.tocsr()
def _reweight_values(self,
doc_term_matrix: sp.csr_matrix) -> sp.csr_matrix:
"""
Re-weight values in a doc-term matrix according to parameters specified
in :class:`Vectorizer` initialization: binary or tf-idf weighting,
sublinear term-frequency, document-normalized weights.
Args:
doc_term_matrix
Returns:
Reweighted doc-term matrix.
"""
# re-weight the local components (term freqs)
if self.tf_type == "binary":
doc_term_matrix.data.fill(1)
elif self.tf_type == "bm25":
if not self.dl_type:
doc_term_matrix.data = (doc_term_matrix.data *
(BM25_K1 + 1.0) /
(BM25_K1 + doc_term_matrix.data))
else:
dls = get_doc_lengths(doc_term_matrix, type_=self.dl_type)
length_norm = (1 - BM25_B) + (BM25_B *
(dls / self._avg_doc_length))
doc_term_matrix = doc_term_matrix.tocoo(copy=False)
doc_term_matrix.data = (
doc_term_matrix.data * (BM25_K1 + 1.0) /
(doc_term_matrix.data +
(BM25_K1 * length_norm[doc_term_matrix.row])))
doc_term_matrix = doc_term_matrix.tocsr(copy=False)
elif self.tf_type == "sqrt":
_ = np.sqrt(doc_term_matrix.data,
doc_term_matrix.data,
casting="unsafe")
elif self.tf_type == "log":
_ = np.log(doc_term_matrix.data,
doc_term_matrix.data,
casting="unsafe")
doc_term_matrix.data += 1.0
elif self.tf_type == "linear":
pass # tfs are already linear
else:
# this should never raise, i'm just being a worrywart
raise ValueError(
errors.value_invalid_msg(
"tf_type", self.tf_type,
{"binary", "bm25", "sqrt", "log", "linear"}))
# apply the global component (idfs), column-wise
if self.idf_type:
doc_term_matrix = doc_term_matrix * self._idf_diag
# apply normalizations, row-wise
# unless we've already handled it for bm25-style tf
if self.dl_type and self.tf_type != "bm25":
n_docs, _ = doc_term_matrix.shape
dls = get_doc_lengths(doc_term_matrix, type_=self.dl_type)
dl_diag = sp.spdiags(1.0 / dls,
diags=0,
m=n_docs,
n=n_docs,
format="csr")
doc_term_matrix = dl_diag * doc_term_matrix
if self.norm is not None:
doc_term_matrix = normalize_mat(doc_term_matrix,
norm=self.norm,
axis=1,
copy=False)
return doc_term_matrix
