fix command line interface of visualization

.gitignore

@@ -5,6 +5,7 @@
 # cmake
 /Makefile
 /src/Makefile
+/cmake-build-*/
 
 # JetBrains
 /.idea

README.md

@@ -78,7 +78,7 @@ Requirements
 
 Generally, GraphVite works on any Linux distribution with CUDA >= 9.2.
 
-The library is compatible with Python 2.7 and 3.5/3.6/3.7.
+The library is compatible with Python 2.7 and 3.6/3.7.
 
 Installation
 ------------
@@ -88,14 +88,16 @@ Installation
 GraphVite can be installed through conda with only one line.
 
 ```bash
-conda install -c milagraph graphvite
+conda install -c milagraph graphvite cudatoolkit=x.x
 ```
 
+where `x.x` is your CUDA version, e.g. 9.2 or 10.0.
+
 If you only need embedding training without evaluation, you can use the following
 alternative with minimal dependencies.
 
 ```bash
-conda install -c milagraph graphvite-mini
+conda install -c milagraph graphvite-mini cudatoolkit=x.x
 ```
 
 ### From Source ###

conda/conda_build_config.yaml

@@ -3,7 +3,6 @@ cxx_compiler_version:
 
 python:
   - 2.7
-  - 3.5
   - 3.6
   - 3.7
 

conda/graphvite-mini/meta.yaml

@@ -8,7 +8,7 @@ source:
 requirements:
   build:
     # cmake
-    - cmake
+    - cmake >=3.12
     - {{ compiler("cxx") }}
     - glog
     - gflags
@@ -22,9 +22,9 @@ requirements:
     - cudatoolkit {{ cudatoolkit }}
     - python {{ python }}
     - pybind11
+    - numpy {{ numpy }}
     - mkl >=2018
     # setup
-    - numpy >=1.11
     - pyyaml
     - easydict
     - six
@@ -41,12 +41,16 @@ requirements:
     - future
 
 build:
-  string: py{{ python|replace(".", "") }}cuda{{ cudatoolkit|replace(".", "") }}
+  string:
+    "py{{ python|replace('.', '') }}\
+    cuda{{ cudatoolkit|replace('.', '') }}\
+    h{{ environ.get('GIT_FULL_HASH')|string|truncate(7, True, '', 0) }}"
 
 test:
   imports:
     - graphvite
 
 about:
   home: https://graphvite.io
-  license: Apache-2.0
+  license: Apache-2.0
+  summary: "A general and high-performance graph embedding system for various applications"

conda/graphvite/meta.yaml

@@ -8,7 +8,7 @@ source:
 requirements:
   build:
     # cmake
-    - cmake
+    - cmake >=3.12
     - {{ compiler("cxx") }}
     - glog
     - gflags
@@ -22,9 +22,9 @@ requirements:
     - cudatoolkit {{ cudatoolkit }}
     - python {{ python }}
     - pybind11
+    - numpy {{ numpy }}
     - mkl >=2018
     # setup
-    - numpy >=1.11
     - pyyaml
     - easydict
     - six
@@ -47,12 +47,16 @@ requirements:
     - nltk
 
 build:
-  string: py{{ python|replace(".", "") }}cuda{{ cudatoolkit|replace(".", "") }}
+  string:
+    "py{{ python|replace('.', '') }}\
+    cuda{{ cudatoolkit|replace('.', '') }}\
+    h{{ environ.get('GIT_FULL_HASH')|string|truncate(7, True, '', 0) }}"
 
 test:
   imports:
     - graphvite
 
 about:
   home: https://graphvite.io
-  license: Apache-2.0
+  license: Apache-2.0
+  summary: "A general and high-performance graph embedding system for various applications"

doc/source/install.rst

@@ -10,15 +10,17 @@ You can install GraphVite from ``conda`` with only one line.
 
 .. code-block:: bash
 
-    conda install -c milagraph graphvite
+    conda install -c milagraph graphvite cudatoolkit=x.x
+
+where ``x.x`` is your CUDA version, e.g. 9.2 or 10.0.
 
 By default, this will install all dependencies, including ``PyTorch`` and
 ``matplotlib``. If you only need embedding training without evaluation, you can take
 the following alternative with minimum dependencies.
 
 .. code-block:: bash
 
-    conda install -c milagraph graphvite-mini
+    conda install -c milagraph graphvite-mini cudatoolkit=x.x
 
 Install from source
 -------------------

include/bind.h

@@ -566,7 +566,7 @@ class pyKnowledgeGraphSolver : public py::class_<graphvite::KnowledgeGraphSolver
             py::arg("model") = "RotatE", py::arg("num_epoch") = 2000, py::arg("resume") = false, py::arg("margin") = 24,
             py::arg("l3_regularization") = 2e-3, py::arg("sample_batch_size") = 2000, py::arg("positive_reuse") = 1,
             py::arg("adversarial_temperature") = 2, py::arg("log_frequency") = 100,
-            "train(model='RotatE', num_epoch=2000, resume=False, margin=24, l3_regulariation=2e-3, "
+            "train(model='RotatE', num_epoch=2000, resume=False, margin=24, l3_regularization=2e-3, "
                   "sample_batch_size=2000, positive_reuse=1, adversarial_temperature=2, log_frequency=100)"
             R"(
             Train knowledge graph embeddings.

include/core/optimizer.h

@@ -192,7 +192,7 @@ class Optimizer {
     template<class Float>
     __device__ inline Float rmsprop_update(Float parameter, Float gradient, Float &moment1, Float weight = 1) const {
         Float regularized = weight * (gradient + weight_decay * parameter);
-        moment1 = alpha * moment1 + (1 - momentum) * regularized * regularized;
+        moment1 = alpha * moment1 + (1 - alpha) * regularized * regularized;
         return lr * regularized / sqrt(moment1 + epsilon);
     }
 

python/graphvite/application/application.py

@@ -32,6 +32,7 @@
 
 logger = logging.getLogger(__name__)
 
+
 class ApplicationMixin(object):
     """
     General interface of graph applications.
@@ -149,10 +150,14 @@ def gpu_map(self, func, settings):
         for i, setting in enumerate(settings):
             new_settings.append(setting + (gpus[i % len(gpus)],))
         settings = new_settings
+
         try:
-            multiprocessing.set_start_method("spawn")
+            start_method = multiprocessing.get_start_method()
+            # if there are other running processes, this could cause leakage of semaphores
+            multiprocessing.set_start_method("spawn", force=True)
             pool = multiprocessing.Pool(len(gpus))
             results = pool.map(func, settings, chunksize=1)
+            multiprocessing.set_start_method(start_method, force=True)
         except AttributeError:
             logger.info("Spawn mode is not supported by multiprocessing. Switch to serial execution.")
             results = list(map(func, settings))

python/graphvite/application/network.py

@@ -23,6 +23,7 @@
 import torch
 from torch import nn
 
+
 class NodeClassifier(nn.Module):
     """
     Node classification network for graphs

python/graphvite/cmd.py

@@ -132,7 +132,7 @@ def load_data(file_name):
     gv.init_logging(logging.WARNING)
 
     app = gap.VisualizationApplication(args.dim, [0])
-    app.load(vectors, perplexity=args.perplexity)
+    app.load(vectors=vectors, perplexity=args.perplexity)
     app.build()
     app.train()
     app.visualization(Y=labels, save_file=args.save)

python/graphvite/dataset.py

@@ -56,6 +56,7 @@
 
 logger = logging.getLogger(__name__)
 
+
 class Dataset(object):
     """
     Graph dataset.

python/graphvite/helper.py

@@ -22,6 +22,7 @@
 from . import lib
 lib.name2dtype = {n: t for t, n in lib.dtype2name.items()}
 
+
 def signature(name, *args):
     strings = [name]
     for arg in args:

python/graphvite/optimizer.py

@@ -24,6 +24,7 @@
 
 module = sys.modules[__name__]
 
+
 class Optimizer(object):
     """
     Optimizer(type=auto, *args, **kwargs)

python/graphvite/util.py

@@ -24,6 +24,7 @@
 
 logger = logging.getLogger(__name__)
 
+
 def recursive_default(obj, default):
     if isinstance(default, dict):
         new_obj = {}
@@ -121,4 +122,5 @@ def wrapper(*args, **kwargs):
 
         return wrapper
 
+
 monitor = Monitor()