Merge pull request #143 from Intron7/update-_pre

Updated pre to run faster

decoupler/method_aucell.py

@@ -8,7 +8,7 @@
 from scipy.sparse import csr_matrix
 
 from numpy.random import default_rng
-from tqdm import tqdm
+from tqdm.auto import tqdm
 
 from .pre import extract, rename_net, filt_min_n, return_data
 

decoupler/method_gsea.py

@@ -12,7 +12,7 @@
 from .pre import extract, rename_net, filt_min_n, return_data
 from .utils import p_adjust_fdr
 
-from tqdm import tqdm
+from tqdm.auto import tqdm
 
 import numba as nb
 

decoupler/method_gsva.py

@@ -13,7 +13,7 @@
 from .pre import extract, rename_net, filt_min_n, return_data
 from .method_gsea import std
 
-from tqdm import tqdm
+from tqdm.auto import tqdm
 
 import numba as nb
 

decoupler/method_mdt.py

@@ -9,7 +9,7 @@
 
 from .pre import extract, match, rename_net, get_net_mat, filt_min_n, return_data
 
-from tqdm import tqdm
+from tqdm.auto import tqdm
 
 
 def check_if_skranger():

decoupler/method_mlm.py

@@ -11,7 +11,7 @@
 
 from scipy import stats
 
-from tqdm import tqdm
+from tqdm.auto import tqdm
 
 import numba as nb
 

decoupler/method_ora.py

@@ -15,7 +15,7 @@
 from .pre import extract, rename_net, filt_min_n, return_data
 from .utils import p_adjust_fdr
 
-from tqdm import tqdm
+from tqdm.auto import tqdm
 
 import numba as nb
 

decoupler/method_udt.py

@@ -9,8 +9,7 @@
 
 from .pre import extract, match, rename_net, get_net_mat, filt_min_n, return_data
 
-from tqdm import tqdm
-
+from tqdm.auto import tqdm
 
 def check_if_sklearn():
     try:

decoupler/method_ulm.py

@@ -11,7 +11,7 @@
 
 from .pre import extract, match, rename_net, get_net_mat, filt_min_n, return_data
 
-from tqdm import tqdm
+from tqdm.auto import tqdm
 
 
 def mat_cov(A, b):

decoupler/method_viper.py

@@ -12,7 +12,8 @@
 
 from .pre import extract, match, rename_net, get_net_mat, filt_min_n, return_data
 
-from tqdm import tqdm
+from tqdm.auto import tqdm
+
 
 import numba as nb
 

decoupler/method_wmean.py

@@ -10,7 +10,7 @@
 from .pre import extract, match, rename_net, get_net_mat, filt_min_n, return_data
 from .method_gsea import std
 
-from tqdm import tqdm
+from tqdm.auto import tqdm
 
 import numba as nb
 

decoupler/method_wsum.py

@@ -10,7 +10,7 @@
 from .pre import extract, match, rename_net, get_net_mat, filt_min_n, return_data
 from .method_gsea import std
 
-from tqdm import tqdm
+from tqdm.auto import tqdm
 
 import numba as nb
 

decoupler/pre.py

@@ -18,7 +18,7 @@ def check_mat(m, r, c, verbose=False):
 
     # Check for empty features
     if type(m) is csr_matrix:
-        msk_features = np.sum(m != 0, axis=0).A1 == 0
+        msk_features = m.getnnz(axis=0) == 0
     else:
         msk_features = np.count_nonzero(m, axis=0) == 0
     n_empty_features = np.sum(msk_features)
@@ -29,16 +29,16 @@ def check_mat(m, r, c, verbose=False):
         m = m[:, ~msk_features]
 
     # Sort features
-    msk = np.argsort(c)
-    m, r, c = m[:, msk], r.astype('U'), c[msk].astype('U')
+    #msk = np.argsort(c)
+    #m, r, c = m[:, msk], r.astype('U'), c[msk].astype('U')
 
     # Check for repeated features
     if np.any(c[1:] == c[:-1]):
         raise ValueError("""mat contains repeated feature names, please make them unique.""")
 
     # Check for empty samples
     if type(m) is csr_matrix:
-        msk_samples = np.sum(m != 0, axis=1).A1 == 0
+        msk_samples = m.getnnz(axis=1) == 0
     else:
         msk_samples = np.count_nonzero(m, axis=1) == 0
     n_empty_samples = np.sum(msk_samples)
@@ -174,9 +174,12 @@ def match(c, r, net):
     # Init empty regX
     regX = np.zeros((c.shape[0], net.shape[1]), dtype=np.float32)
 
-    # Match genes from mat, else are 0s
-    idxs = np.searchsorted(c, r)
-    regX[idxs] = net
+    # Create an index array for rows of c corresponding to r
+    c_dict = {gene: i for i, gene in enumerate(c)}
+    idxs = [c_dict[gene] for gene in r if gene in c_dict]
+
+    # Populate regX using advanced indexing
+    regX[idxs, :] = net[: len(idxs), :]
 
     return regX
 

decoupler/utils_benchmark.py

@@ -332,7 +332,7 @@ def format_acts_grts(res, obs, groupby, use_pval):
     grts = build_grts_mat(obs, exps, srcs)
 
     # Match to same srcs between acts and grts
-    grts = match(srcs, grts.columns, grts.T).T
+    grts = match(srcs, grts.columns, grts.T.values).T
 
     # Build msks tensor
     msks, grpbys, grps = build_msks_tensor(obs, groupby)