Put graphviz as optional requirement

Enable LNN installation without plotting ability, plotting modules can be installed via extra requirements.

.github/workflows/build.yml

@@ -22,8 +22,6 @@ jobs:
         uses: actions/setup-python@v2
         with:
           python-version: ${{ matrix.python-version }}
-      - name: Install graphviz
-        run: sudo apt-get update && sudo apt-get install -y graphviz graphviz-dev
       - name: Upgrade pip
         run: python -m pip install --upgrade pip setuptools wheel
       - name: Install dependencies

README.md

@@ -21,10 +21,16 @@ properties of both neural nets (learning) and symbolic logic (knowledge and reas
 ## Quickstart
 
 To install the LNN:
+1. Run: 
+   ```
+   pip install git+https://github.com/IBM/LNN.git
+   ```
+
+To install the LNN with graph plot support:
 1. Install [GraphViz](https://www.graphviz.org/download/)
 2. Run: 
    ```
-   pip install git+https://github.com/IBM/LNN.git
+   pip install git+https://github.com/IBM/LNN.git#egg=lnn"[plot]"
    ```
 
 ## Documentation

lnn/_utils.py

@@ -8,7 +8,6 @@
 from typing import Union, TypeVar, Tuple
 
 import torch
-import torchviz
 import numpy as np
 
 from . import _exceptions
@@ -177,6 +176,8 @@ def dict_rekey(d, old_key, new_key) -> None:
 
 
 def plot_autograd(model: Model, loss: torch.Tensor, **kwds) -> None:
+    import torchviz
+
     params = model.named_parameters()
     torchviz.make_dot(
         loss,

lnn/utils.py

@@ -10,7 +10,6 @@
 
 import numpy as np
 import networkx as nx
-import matplotlib.pyplot as plt
 
 from . import _utils
 from .constants import Fact
@@ -78,6 +77,8 @@ def predicate_truth_table(*args: str, arity: int, model, states=None):
 
 
 def plot_graph(self, **kwds) -> None:
+    import matplotlib.pyplot as plt
+
     labels = {node: f"{node.__class__.__name__}\n{node}" for node in self.graph}
 
     options = {
@@ -97,6 +98,8 @@ def plot_graph(self, **kwds) -> None:
 
 
 def plot_loss(total_loss, losses) -> None:
+    import matplotlib.pyplot as plt
+
     loss, cummulative_loss = total_loss
     fig, axs = plt.subplots(1, 2)
     fig.suptitle("Model Loss")
@@ -113,6 +116,8 @@ def plot_loss(total_loss, losses) -> None:
 
 
 def plot_params(self: Model) -> None:
+    import matplotlib.pyplot as plt
+
     legend = []
     for node in self.nodes:
         if hasattr(self[node], "parameter_history"):

requirements.txt

@@ -1,8 +1,4 @@
 networkx==2.5.1
-torch==1.9.0
-matplotlib==3.3.3
-pygraphviz==1.6
+torch==1.11.0
 tqdm==4.50.2
-numpy~=1.21.0
-setuptools~=52.0.0
-torchviz==0.0.2
+numpy==1.21.0

requirements_plot.txt

@@ -0,0 +1,3 @@
+pygraphviz==1.6
+torchviz==0.0.2
+matplotlib==3.3.3

requirements_test.txt

@@ -0,0 +1 @@
+pytest==7.1.0

setup.py

@@ -8,6 +8,10 @@
 import pathlib
 
 
+def parse_requirements(filename):
+    return pathlib.Path(filename).read_text().replace("==", ">=").split("\n")
+
+
 setuptools.setup(
     name="lnn",
     version="1.0",
@@ -17,9 +21,10 @@
     long_description_content_type="text/markdown",
     url="https://github.com/IBM/LNN",
     packages=setuptools.find_packages(),
-    install_requires=pathlib.Path("requirements.txt").read_text().replace("==", ">="),
+    install_requires=parse_requirements("requirements.txt"),
     extras_require={
-        "test": ["pytest"],
+        "test": parse_requirements("requirements_test.txt"),
+        "plot": parse_requirements("requirements_plot.txt"),
     },
     classifiers=[
         "Programming Language :: Python :: 3",