Merge pull request #17 from soo-h/master

merge updates for v1.1.2

MuRaL/nn_utils.py

@@ -123,7 +123,7 @@ def run_time_view_model_predict_m(model, dataloader, criterion, device, n_class,
     total_loss = 0
     batch_count = 0
     get_batch_time_recode = 0
-    get_batch_train_recode = 0
+    get_batch_predict_recode = 0
     step_time = time.time()
     get_batch_time = time.time()
 
@@ -136,7 +136,7 @@ def run_time_view_model_predict_m(model, dataloader, criterion, device, n_class,
                 print("get 500 batch used time: ", get_batch_time_recode)
                 get_batch_time_recode = 0
 
-            batch_train_time = time.time()
+            batch_predict_time = time.time()
             cat_x = cat_x.to(device)
             cont_x = cont_x.to(device)
             distal_x = distal_x.to(device)
@@ -155,10 +155,10 @@ def run_time_view_model_predict_m(model, dataloader, criterion, device, n_class,
                 print(f"Batch Number: {batch_count}; Mean Time of 500 batch: {(time.time()-step_time)}")
                 step_time = time.time()
 
-            get_batch_train_recode += time.time() - batch_train_time
+            get_batch_predict_recode += time.time() - batch_predict_time
             if batch_count % 500 == 0:
-                print("training 500 batch used time:", get_batch_train_recode)
-                get_batch_train_recode = 0
+                print("training 500 batch used time:", get_batch_predict_recode)
+                get_batch_predict_recode = 0
             get_batch_time = time.time()
 
             if device == torch.device('cpu'):

MuRaL/run_predict.py

@@ -113,12 +113,6 @@ def parse_arguments(parser):
                           between RAM memory and preprocessing speed. It is recommended to use 300k.
                           Default: 300000.""" ).strip())
 
-    optional.add_argument('--sampled_segments', type=int, metavar='INT', default=[10], nargs='+',
-                          help=textwrap.dedent("""
-                          Number of segments chosen for generating samples for batches in DataLoader.
-                          Default: 10.
-                          """ ).strip())
-
     optional.add_argument('--pred_batch_size', metavar='INT', default=16, 
                           help=textwrap.dedent("""
                           Size of mini batches for prediction. Default: 16.
@@ -208,7 +202,7 @@ def main():
     ref_genome= args.ref_genome
 
     pred_batch_size = args.pred_batch_size
-    sampled_segments = args.sampled_segments
+    sampled_segments = 1
     # Output file path
     pred_file = args.pred_file
 

MuRaL/run_train_TL_raytune.py

@@ -441,7 +441,9 @@ def main():
 
         if not sampled_segments:
             sampled_segments = args.sampled_segments = para_read_from_config('sampled_segments', config)
-
+        else:
+            sampled_segments = args.sampled_segments[0]
+
         args.seq_only = config['seq_only']
 
 

MuRaL/run_train_raytune.py

@@ -209,7 +209,7 @@ def parse_arguments(parser):
                           Default: 0.25.
                            """ ).strip())
 
-	learn_args.add_argument('--segment_center', type=int, metavar='INT', default=300000, 
+    learn_args.add_argument('--segment_center', type=int, metavar='INT', default=300000, 
                           help=textwrap.dedent("""
                           The maximum encoding unit of the sequence. It affects trade-off 
                           between RAM memory and preprocessing speed. It is recommended to use 300k.

dirichlet_python/dirichletcal/calib/multinomial.py

@@ -274,7 +274,7 @@ def _newton_update(weights_0, X, XX_T, target, k, method_, maxiter=int(1024),
             updates = gradient / hessian
         else:
             try:
-                inverse = scipy.linalg.pinv2(hessian)
+                inverse = scipy.linalg.pinv(hessian)
                 updates = np.matmul(inverse, gradient)
             except (raw_np.linalg.LinAlgError, ValueError) as err:
                 logging.error(err)

dirichlet_python/setup.py

@@ -29,11 +29,11 @@
     ],
     python_requires='>=3.6',
     install_requires = [
-        'numpy>=1.14.2'
-        'scipy>=1.0.0'
-        'scikit-learn>=0.19.1'
-        'jax'
-        'jaxlib'
+        'numpy>=1.14.2',
+        'scipy>=1.0.0',
+        'scikit-learn>=0.19.1',
+        'jax',
+        'jaxlib',
         'autograd'
     ]
 )

environment.yml

@@ -1,11 +1,13 @@
 channels:
+  - pytorch
   - bioconda
   - conda-forge
   - defaults
 dependencies:
   - python=3.8.5
   - pip=22.0.4
-  - numpy=1.21.2
+  - numpy=1.23.5
+  - pandas=1.4.1
   - cudatoolkit=11.3
   - pytorch=1.10.2
   - pybigwig=0.3.17
@@ -24,3 +26,4 @@ dependencies:
     - jaxlib==0.4.13
     - autograd==1.6.2
     - protobuf==3.20.0
+    - setuptools==69.0.2

environment_cpu.yml

@@ -1,11 +1,13 @@
 channels:
+  - pytorch
   - bioconda
   - conda-forge
   - defaults
 dependencies:
   - python=3.8.5
   - pip=22.0.4
-  - numpy=1.21.2
+  - numpy=1.23.5
+  - pandas=1.4.1
   - pytorch=1.10.2
   - cpuonly=2.0
   - pybigwig=0.3.17
@@ -24,3 +26,4 @@ dependencies:
     - jaxlib==0.4.13
     - autograd==1.6.2
     - protobuf==3.20.0
+    - setuptools==69.0.2

setup.py

@@ -4,8 +4,12 @@
 
 from setuptools import setup, find_packages
 
-import os
-os.system('bash dirichlet_install.sh')
+import subprocess
+try:
+    subprocess.check_call(['pip', 'install', '.'], cwd='dirichlet_python')
+except subprocess.CalledProcessError as e:
+    print(f"Error installing dirichlet_python: {e}")
+    exit(1)
 
 def get_version():
     try: