Merge pull request #54 from MPI-Dortmund/random-port

Use random port for distributed processing

.github/workflows/checks_tests_deploy.yml

@@ -45,7 +45,7 @@ jobs:
     - name: Install dependencies
       run: |
         python -m pip install --upgrade pip
-        pip install setuptools setuptools_scm pylint tox tox-gh-actions pytest pytest-coverage twine build
+        pip install setuptools setuptools_scm pylint==2.17.7 tox tox-gh-actions pytest pytest-coverage twine build
         pip install .
     - name: Debug Info
       run: |

scripts/molmapsbbox.py

@@ -1,19 +1,22 @@
-from skimage.measure import label, regionprops
-import numpy as np
+import argparse
+import json
+import os
 from typing import List
+
 import mrcfile
+import numpy as np
+import skimage.measure as skm
+
 from tomotwin.modules.common.preprocess import label_filename
-import argparse
-import os
-import json
+
 
 def binarize(vol):
     t = np.max(vol) - (np.max(vol) - np.min(vol)) * 0.9
     return vol > t
 
 def boxsize(vol):
-    lbl = label(vol, background=0)
-    regs = regionprops(lbl)
+    lbl = skm.label(vol, background=0)
+    regs = skm.regionprops(lbl)
     size = -1
     for reg in regs:
         cand_size = np.max([np.abs(reg.bbox[i] - reg.bbox[i+3]) for i in range(int(len(reg.bbox)/2))])

tests/test_embed_main.py

@@ -83,6 +83,10 @@ def test_embed_main_real_subvol_torchembeddor(self):
     )
     @pytest.mark.skipif(torch.cuda.is_available() == False, reason="Skipped because CUDA is not available")
     def test_embed_main_real_subvol_distributedtorchembeddor(self):
+        import random
+        os.environ['MASTER_ADDR'] = '127.0.0.1'
+        os.environ['MASTER_PORT'] = '29' + str(random.randint(1, 500)).zfill(3)
+
         from tomotwin.embed_main import run as embed_main_func
 
         tomo = np.random.randn(37, 37, 37).astype(np.float32)

tomotwin/embed_main.py

@@ -377,6 +377,7 @@
 import glob
 import hashlib
 import os
+import random
 import resource
 import sys
 from typing import List
@@ -605,6 +606,8 @@ def run_distr(config, world_size: int):
             f"Your user limit ('ulimit -n') is too low ({limit[0]}). Please run 'ulimit -n 65000' before running tomotwin_embed.")
         sys.exit(1)
     print(f"Found {world_size} GPU(s). Start DDP + Compiling.")
+    os.environ['MASTER_ADDR'] = '127.0.0.1'
+    os.environ['MASTER_PORT'] = '29' + str(random.randint(1, 500)).zfill(3)
     mp.spawn(
         run,
         args=([config, world_size]),

tomotwin/modules/common/io/mrc_format.py

@@ -381,8 +381,10 @@
 class MrcFormat:
     @staticmethod
     def read(pth: str) -> np.array:
+        vol = None
         try:
-            vol = mrcfile.open(pth, mode="r", permissive=True)
+            with mrcfile.open(pth, mode="r", permissive=True) as mrc:
+                vol = mrc.data
         except ValueError as e:
             raise Exception(f"Failed reading {pth}") from e
-        return vol.data
+        return vol

tomotwin/modules/inference/embedor.py

@@ -375,7 +375,6 @@
   defined by the Mozilla Public License, v. 2.0.
 """
 
-import os
 from abc import ABC, abstractmethod
 
 import numpy as np
@@ -522,9 +521,6 @@ def __init__(
             workers: int = 0,
     ) -> None:
         """Inits the embedor"""
-
-        os.environ['MASTER_ADDR'] = '127.0.0.1'
-        os.environ['MASTER_PORT'] = '29500'
         tdist.init_process_group(backend='gloo', rank=rank, world_size=world_size)
         tdist.barrier()
         self.rank = rank

tomotwin/modules/networks/SiameseNet3D.py

@@ -524,8 +524,7 @@ def forward(self, inputtensor):
             Forward pass through the network
             :param inputtensor: Input tensor
             """
-
-            inputtensor = nn.functional.pad(inputtensor, (1, 2, 1, 2, 1, 2))
+            inputtensor = F.pad(inputtensor, (1, 2, 1, 2, 1, 2))
 
             out = self.conv_layer0(inputtensor)
             out = self.max_pooling(out)