Drop support for Python 3.5

.appveyor.yml

@@ -4,7 +4,7 @@ image: Visual Studio 2019
 
 environment:
   MODE: test
-  PYTHON_VERSION: C:\Python36-x64
+  PYTHON_VERSION: C:\Python37-x64
   NUMPY_VERSION: numpy
   TF_VERSION: tensorflow
   NENGO_VERSION: nengo[tests]
@@ -18,7 +18,7 @@ environment:
     - NENGO_VERSION: nengo[tests]==3.0.0
       TF_VERSION: tensorflow==2.2.0
       NUMPY_VERSION: numpy==1.16.0
-      PYTHON_VERSION: C:\Python35-x64
+      PYTHON_VERSION: C:\Python36-x64
 
 init:
   - set PATH=%PYTHON_VERSION%;%PYTHON_VERSION%\Scripts;%PATH%

.nengobones.yml

@@ -231,7 +231,7 @@ ci_scripts:
   - template: deploy
 
 travis_yml:
-  python: 3.6
+  python: 3.7
   global_vars:
     NUMPY_VERSION: numpy>=1.16.0
     TF_VERSION: tensorflow
@@ -256,7 +256,7 @@ travis_yml:
         NENGO_VERSION: nengo[tests]==3.0.0
         TF_VERSION: tensorflow==2.2.0
         NUMPY_VERSION: numpy==1.16.0
-      python: 3.5
+      python: 3.6
     - script: test-coverage
       test_args: --graph-mode
     - stage: advanced
@@ -315,7 +315,6 @@ setup_py:
       - "Operating System :: Microsoft :: Windows"
       - "Operating System :: POSIX :: Linux"
       - "Programming Language :: Python"
-      - "Programming Language :: Python :: 3.5"
       - "Programming Language :: Python :: 3.6"
       - "Programming Language :: Python :: 3.7"
       - "Programming Language :: Python :: 3.8"

.templates/setup.py.template

@@ -19,7 +19,7 @@ else:
     # environment, so we can't just look up the tensorflow version in the current
     # environment. but the pip package will be in the isolated sys.path, so we can use
     # that to look up the site-packages directory of the original environment.
-    target_path = os.path.join("site-packages", "pip")
+    target_path = str(pathlib.Path("site-packages", "pip"))
     for path in sys.path:
         if target_path in path:
             source_path = [path[: path.index("pip")]]
@@ -44,7 +44,7 @@ else:
 install_req = [
     "nengo>=3.0.0",
     "numpy>=1.16.0",
-    "%s>=2.2.0" % tf_req,
+    "{}>=2.2.0".format(tf_req),
     "jinja2>=2.10.1",
     "packaging>=20.0",
     "progressbar2>=3.39.0",

.travis.yml

@@ -1,7 +1,7 @@
 # Automatically generated by nengo-bones, do not edit this file directly
 
 language: python
-python: 3.6
+python: 3.7
 notifications:
   email:
     on_success: change
@@ -55,7 +55,7 @@ jobs:
       TF_VERSION="tensorflow==2.2.0"
       NUMPY_VERSION="numpy==1.16.0"
       SCRIPT="test-coverage"
-    python: 3.5
+    python: 3.6
   -
     env:
       SCRIPT="test-coverage"
@@ -111,11 +111,7 @@ before_install:
   # install/run nengo-bones
   - pip install git+https://github.com/nengo/nengo-bones#egg=nengo-bones
   - bones-generate --output-dir .ci ci-scripts
-  - if [[ "$TRAVIS_PYTHON_VERSION" < "3.6" ]]; then
-        echo "Skipping bones-check because Python $TRAVIS_PYTHON_VERSION < 3.6";
-    else
-        bones-check --verbose;
-    fi
+  - bones-check --verbose
   # display environment info
   - pip freeze
 

CHANGES.rst

@@ -25,6 +25,13 @@ Release history
 
 *Compatible with TensorFlow 2.2.0 - 2.4.0*
 
+**Removed**
+
+- Dropped support for Python 3.5 (which reached its end of life in September 2020).
+  (`#184`_)
+
+.. _#184: https://github.com/nengo/nengo-dl/pull/184
+
 3.4.0 (November 26, 2020)
 -------------------------
 

docs/conf.py

@@ -2,7 +2,7 @@
 #
 # Automatically generated by nengo-bones, do not edit this file directly
 
-import os
+import pathlib
 
 import nengo_dl
 
@@ -100,11 +100,11 @@
 templates_path = ["_templates"]
 html_static_path = ["_static"]
 html_theme = "nengo_sphinx_theme"
-html_title = "NengoDL {0} docs".format(release)
+html_title = f"NengoDL {release} docs"
 htmlhelp_basename = "NengoDL"
 html_last_updated_fmt = ""  # Default output format (suppressed)
 html_show_sphinx = False
-html_favicon = os.path.join("_static", "favicon.ico")
+html_favicon = str(pathlib.Path("_static", "favicon.ico"))
 html_theme_options = {
     "nengo_logo": "nengo-dl-full-light.svg",
     "nengo_logo_color": "#ff6600",

docs/config.rst

@@ -300,3 +300,33 @@ setting ``nengo_dl.Simulator(..., unroll_simulation=x)``. This will explicitly b
 ``x`` timesteps into the model (without using a loop).  So if we use
 ``unroll_simulation=x`` and ``use_loop=False``, then the simulation will always run
 for exactly ``x`` timesteps.
+
+learning_phase
+--------------
+
+NengoDL (and Keras in general) can run models in two different modes, or "phases":
+training and inference. Typically the mode is set automatically based on the function
+used to execute the model; ``sim.fit`` will run in training mode, and all other
+functions (e.g., ``sim.predict``, ``sim.run``, and ``sim.evaluate``) run in inference
+mode.
+
+The most important way that the learning phase affects a NengoDL model is that it
+controls whether spiking neurons are simulated in spiking or non-spiking mode.
+Normally during training spiking neurons will automatically swap their behaviour to
+a non-spiking equivalent, and use the spiking behaviour during inference.
+
+However, sometimes it can be useful to override this default behaviour. For example,
+we might want to evaluate a spiking model in training mode, to get more insight into
+how it is behaving during training. That is the role of the ``learning_phase``
+config option. It can be specified during Network construction to make a network
+that will *always* run in training/inference mode, regardless of what function is
+being called.  For example:
+
+.. testcode::
+
+    with nengo.Network() as net:
+        nengo_dl.configure_settings(learning_phase=True)
+
+        # this ensemble will always use the "training" mode of LIF
+        # (equivalent to LIFRate)
+        ens = nengo.Ensemble(10, 1, neuron_type=nengo.LIF())

docs/examples/keras-to-snn.ipynb

@@ -260,7 +260,7 @@
     "    # last timestep\n",
     "    predictions = np.argmax(data[nengo_output][:, -1], axis=-1)\n",
     "    accuracy = (predictions == test_labels[:n_test, 0, 0]).mean()\n",
-    "    print(\"Test accuracy: %.2f%%\" % (100 * accuracy))\n",
+    "    print(f\"Test accuracy: {100 * accuracy:.2f}%\")\n",
     "\n",
     "    # plot the results\n",
     "    for ii in range(3):\n",
@@ -282,7 +282,8 @@
     "            plt.ylabel(\"Firing rates (Hz)\")\n",
     "        plt.xlabel(\"Timestep\")\n",
     "        plt.title(\n",
-    "            \"Neural activities (conv0 mean=%dHz max=%dHz)\" % (rates.mean(), rates.max())\n",
+    "            f\"Neural activities (conv0 mean={rates.mean():.1f} Hz, \"\n",
+    "            f\"max={rates.max():.1f} Hz)\"\n",
     "        )\n",
     "        plt.plot(scaled_data)\n",
     "\n",
@@ -387,7 +388,7 @@
    "outputs": [],
    "source": [
     "for s in [0.001, 0.005, 0.01]:\n",
-    "    print(\"Synapse=%.3f\" % s)\n",
+    "    print(f\"Synapse={s:.3f}\")\n",
     "    run_network(\n",
     "        activation=nengo.SpikingRectifiedLinear(),\n",
     "        n_steps=60,\n",
@@ -445,7 +446,7 @@
    "outputs": [],
    "source": [
     "for scale in [2, 5, 10]:\n",
-    "    print(\"Scale=%d\" % scale)\n",
+    "    print(f\"Scale={scale}\")\n",
     "    run_network(\n",
     "        activation=nengo.SpikingRectifiedLinear(),\n",
     "        scale_firing_rates=scale,\n",

docs/examples/spa-memory.ipynb

@@ -77,7 +77,7 @@
     "\n",
     "    # iterate through examples to be generated, fill arrays\n",
     "    for n in range(n_inputs):\n",
-    "        name = \"SP_%d\" % n\n",
+    "        name = f\"SP_{n}\"\n",
     "        vocab.add(name, vocab.create_pointer())\n",
     "\n",
     "        # create inputs and target memory for first pair\n",
@@ -163,7 +163,7 @@
     "def plot_memory_example(plot_sim, vocab, example_input=0):\n",
     "    plt.figure(figsize=(8, 8))\n",
     "\n",
-    "    name = \"SP_%d\" % example_input\n",
+    "    name = f\"SP_{example_input}\"\n",
     "\n",
     "    plt.subplot(3, 1, 1)\n",
     "    plt.plot(\n",
@@ -351,8 +351,8 @@
     "\n",
     "    # iterate through examples to be generated, fill arrays\n",
     "    for n in range(n_items):\n",
-    "        role_names = [\"ROLE_%d_%d\" % (n, i) for i in range(pairs_per_item)]\n",
-    "        filler_names = [\"FILLER_%d_%d\" % (n, i) for i in range(pairs_per_item)]\n",
+    "        role_names = [f\"ROLE_{n}_{i}\" for i in range(pairs_per_item)]\n",
+    "        filler_names = [f\"FILLER_{n}_{i}\" for i in range(pairs_per_item)]\n",
     "\n",
     "        # each role/filler pair is presented for presentation_time seconds\n",
     "        for i in range(pairs_per_item):\n",
@@ -361,7 +361,7 @@
     "                filler_names[i]\n",
     "            ).v\n",
     "            binding[n, i * int_steps : (i + 1) * int_steps] = vocab.parse(\n",
-    "                \"%s*%s\" % (role_names[i], filler_names[i])\n",
+    "                f\"{role_names[i]}*{filler_names[i]}\"\n",
     "            ).v\n",
     "\n",
     "        # randomly select a cue\n",

docs/examples/spa-retrieval.ipynb

@@ -79,14 +79,14 @@
     "\n",
     "    # iterate through all of the examples to be generated\n",
     "    for n in range(n_items):\n",
-    "        role_names = [\"ROLE_%d_%d\" % (n, i) for i in range(pairs_per_item)]\n",
-    "        filler_names = [\"FILLER_%d_%d\" % (n, i) for i in range(pairs_per_item)]\n",
+    "        role_names = [f\"ROLE_{n}_{i}\" for i in range(pairs_per_item)]\n",
+    "        filler_names = [f\"FILLER_{n}_{i}\" for i in range(pairs_per_item)]\n",
     "\n",
     "        # create key for the 'trace' of bound pairs (i.e. a\n",
     "        # structured semantic pointer)\n",
     "        trace_key = \"TRACE_\" + str(n)\n",
     "        trace_ptr = vocab.parse(\n",
-    "            \"+\".join(\"%s * %s\" % (x, y) for x, y in zip(role_names, filler_names))\n",
+    "            \"+\".join(f\"{x} * {y}\" for x, y in zip(role_names, filler_names))\n",
     "        )\n",
     "        trace_ptr.normalize()\n",
     "        vocab.add(trace_key, trace_ptr)\n",
@@ -96,8 +96,8 @@
     "\n",
     "        # fill array elements correspond to this example\n",
     "        traces[n, 0, :] = vocab[trace_key].v\n",
-    "        cues[n, 0, :] = vocab[\"ROLE_%d_%d\" % (n, cue_idx)].v\n",
-    "        targets[n, 0, :] = vocab[\"FILLER_%d_%d\" % (n, cue_idx)].v\n",
+    "        cues[n, 0, :] = vocab[f\"ROLE_{n}_{cue_idx}\"].v\n",
+    "        targets[n, 0, :] = vocab[f\"FILLER_{n}_{cue_idx}\"].v\n",
     "\n",
     "    return traces, cues, targets, vocab"
    ]

docs/installation.rst

@@ -13,7 +13,7 @@ That's it!
 
 Requirements
 ------------
-NengoDL works with Python 3.5 or later.  ``pip`` will do its best to install
+NengoDL works with Python 3.6 or later.  ``pip`` will do its best to install
 all of NengoDL's requirements when it installs NengoDL.  However, if anything
 goes wrong during this process you can install the requirements manually and
 then try to ``pip install nengo-dl`` again.

nengo_dl/__init__.py

@@ -5,32 +5,6 @@
 __license__ = "Free for non-commercial use; see LICENSE.rst"
 from nengo_dl.version import version as __version__
 
-# check python version
-import sys
-
-if sys.version_info < (3, 5):
-    raise ImportError(
-        """
-You are running Python version %s with NengoDL version %s. NengoDL requires
-at least Python 3.5.
-
-The fact that this version was installed on your system probably means that you
-are using an older version of pip; you should consider upgrading with
-
- $ pip install pip setuptools --upgrade
-
-There are two options for getting NengoDL working:
-
-- Upgrade to Python >= 3.5
-
-- Install an older version of NengoDL:
-
- $ pip install 'nengo-dl<2.0'
-"""
-        % (sys.version, __version__)
-    )
-del sys
-
 # filter out "INFO" level log messages
 import os
 

nengo_dl/benchmarks.py

@@ -747,8 +747,8 @@ def build(obj, benchmark, dimensions, neurons_per_d, neuron_type, kwarg):
 
     # build benchmark and add to context for chaining
     print(
-        "Building %s with %s"
-        % (nengo_dl.utils.function_name(benchmark, sanitize=False), kwargs)
+        f"Building {nengo_dl.utils.function_name(benchmark, sanitize=False)} "
+        f"with {kwargs}"
     )
 
     obj["net"] = benchmark(**kwargs)

nengo_dl/builder.py

@@ -38,7 +38,7 @@ def __init__(self, plan):
         for ops in self.plan:
             if type(ops[0]) not in Builder.builders:
                 raise BuildError(
-                    "No registered builder for operators of type %r" % type(ops[0])
+                    f"No registered builder for operators of type {type(ops[0])!r}"
                 )
             self.op_builds[ops] = Builder.builders[type(ops[0])](ops)
 
@@ -157,7 +157,7 @@ def register_builder(build_class):
 
             if nengo_op in cls.builders:
                 warnings.warn(
-                    "Operator '%s' already has a builder. Overwriting." % nengo_op
+                    f"Operator '{nengo_op}' already has a builder. Overwriting."
                 )
 
             cls.builders[nengo_op] = build_class

nengo_dl/callbacks.py

@@ -47,7 +47,7 @@ def __init__(self, log_dir, sim, objects):
         self.sim = sim
 
         with contextlib.suppress() if compat.eager_enabled() else context.eager_mode():
-            self.writer = tf.summary.create_file_writer(log_dir)
+            self.writer = tf.summary.create_file_writer(str(log_dir))
 
         self.summaries = []
         for obj in objects:
@@ -56,26 +56,26 @@ def __init__(self, log_dir, sim, objects):
             ):
                 if isinstance(obj, nengo.Ensemble):
                     param = "encoders"
-                    name = "Ensemble_%s" % obj.label
+                    name = f"Ensemble_{obj.label}"
                 elif isinstance(obj, nengo.ensemble.Neurons):
                     param = "bias"
-                    name = "Ensemble.neurons_%s" % obj.ensemble.label
+                    name = f"Ensemble.neurons_{obj.ensemble.label}"
                 elif isinstance(obj, nengo.Connection):
                     if not compat.conn_has_weights(obj):
                         raise ValidationError(
-                            "Connection '%s' does not have any weights to log" % obj,
+                            f"Connection '{obj}' does not have any weights to log",
                             "objects",
                         )
                     param = "weights"
-                    name = "Connection_%s" % obj.label
+                    name = f"Connection_{obj.label}"
 
                 self.summaries.append(
-                    (utils.sanitize_name("%s_%s" % (name, param)), obj, param)
+                    (utils.sanitize_name(f"{name}_{param}"), obj, param)
                 )
             else:
                 raise ValidationError(
-                    "Unknown summary object %s; should be an Ensemble, Neurons, or "
-                    "Connection" % obj,
+                    f"Unknown summary object {obj}; should be an Ensemble, Neurons, or "
+                    "Connection",
                     "objects",
                 )