Merge pull request #478 from BindsNET/hananel

Update BindsNET to PyTorch 1.8.1

README.md

@@ -1,12 +1,3 @@
-# Important:
-
-If you are using BindsNET for your research, please consider answering a short survey on the value of a standardized format that could be used for the exchange of computational models in neuroscience, cognitive science, and computer science.
-
-[Survey link](https://docs.google.com/forms/d/e/1FAIpQLSfOZSyoWL4J_MfHTIq5Jx0dnJSVHNFH7UP3AHG_x7ddlMvBPA/viewform?usp=sf_link)
-
-Data from this survey will be used exclusively for our own internal purposes, and will not be shared beyond our small group other than summary form for any purpose. The survey is anonymous unless you chose to add an email address for follow-up correspondence. 
-
-
 <p align="center"><img width="25%" src="docs/logo.png"/></p>
 
 A Python package used for simulating spiking neural networks (SNNs) on CPUs or GPUs using [PyTorch](http://pytorch.org/) `Tensor` functionality.

bindsnet/analysis/plotting.py

@@ -228,6 +228,7 @@ def plot_weights(
             save = a[0] + a[1]
 
         plt.savefig(save, bbox_inches="tight")
+        plt.savefig(a[0] + ".png", bbox_inches="tight")
 
         plt.close(fig)
         plt.ion()
@@ -407,6 +408,13 @@ def plot_assignments(
     if save is not None:
         plt.ioff()
 
+        a = save.split(".")
+        if len(a) == 2:
+            save = a[0] + ".1." + a[1]
+        else:
+            a[1] = "." + str(1 + int(a[1])) + ".png"
+            save = a[0] + a[1]
+
         fig, ax = plt.subplots(figsize=figsize)
         ax.set_title("Categorical assignments")
 
@@ -440,7 +448,7 @@ def plot_assignments(
         plt.close()
 
         plt.ion()
-        return im
+        return im, save
     else:
         if not im:
             fig, ax = plt.subplots(figsize=figsize)
@@ -749,8 +757,7 @@ def plot_voltages(
                         v[1]
                         .cpu()
                         .numpy()[
-                            time[0] : time[1],
-                            n_neurons[v[0]][0] : n_neurons[v[0]][1],
+                            time[0] : time[1], n_neurons[v[0]][0] : n_neurons[v[0]][1]
                         ]
                     )
                     if thresholds is not None and thresholds[v[0]].size() == torch.Size(
@@ -764,8 +771,7 @@ def plot_voltages(
                         v[1]
                         .cpu()
                         .numpy()[
-                            time[0] : time[1],
-                            n_neurons[v[0]][0] : n_neurons[v[0]][1],
+                            time[0] : time[1], n_neurons[v[0]][0] : n_neurons[v[0]][1]
                         ]
                         .T,
                         cmap=cmap,

bindsnet/encoding/encodings.py

@@ -141,7 +141,7 @@ def poisson(
 
         # Create Poisson distribution and sample inter-spike intervals
         # (incrementing by 1 to avoid zero intervals).
-        dist = torch.distributions.Poisson(rate=rate)
+        dist = torch.distributions.Poisson(rate=rate, validate_args=False)
         intervals = dist.sample(sample_shape=torch.Size([time + 1]))
         intervals[:, datum != 0] += (intervals[:, datum != 0] == 0).float()
 

bindsnet/evaluation/evaluation.py

@@ -207,7 +207,7 @@ def ngram(
 
         predictions.append(torch.argmax(score))
 
-    return torch.tensor(predictions).long()
+    return torch.tensor(predictions, device=spikes.device).long()
 
 
 def update_ngram_scores(
@@ -246,7 +246,7 @@ def update_ngram_scores(
         for order in zip(*(fire_order[k:] for k in range(n))):
             for sequence in product(*order):
                 if sequence not in ngram_scores:
-                    ngram_scores[sequence] = torch.zeros(n_labels)
+                    ngram_scores[sequence] = torch.zeros(n_labels, device=spikes.device)
 
                 ngram_scores[sequence][int(labels[i])] += 1
 

bindsnet/network/nodes.py

@@ -101,7 +101,7 @@ def forward(self, x: torch.Tensor) -> None:
             if self.traces_additive:
                 self.x += self.trace_scale * self.s.float()
             else:
-                self.x.masked_fill_(self.s, 1)
+                self.x.masked_fill_(self.s.bool(), 1)
 
         if self.sum_input:
             # Add current input to running sum.

bindsnet/network/topology.py

@@ -794,12 +794,12 @@ def __init__(
             )
             if self.wmin == -np.inf or self.wmax == np.inf:
                 v = torch.clamp(
-                    torch.rand(*source.shape, *target.shape)[i.byte()],
+                    torch.rand(*source.shape, *target.shape)[i.bool()],
                     self.wmin,
                     self.wmax,
                 )
             else:
-                v = self.wmin + torch.rand(*source.shape, *target.shape)[i.byte()] * (
+                v = self.wmin + torch.rand(*source.shape, *target.shape)[i.bool()] * (
                     self.wmax - self.wmin
                 )
             w = torch.sparse.FloatTensor(i.nonzero().t(), v)

examples/mnist/SOM_LM-SNNs.py

@@ -77,7 +77,7 @@
 
 # Determines number of workers to use
 if n_workers == -1:
-    n_workers = torch.cuda.is_available() * 4 * torch.cuda.device_count()
+    n_workers = 0  # torch.cuda.is_available() * 4 * torch.cuda.device_count()
 
 n_sqrt = int(np.ceil(np.sqrt(n_neurons)))
 start_intensity = intensity

examples/mnist/batch_eth_mnist.py

@@ -49,7 +49,7 @@
 parser.add_argument("--test", dest="train", action="store_false")
 parser.add_argument("--plot", dest="plot", action="store_true")
 parser.add_argument("--gpu", dest="gpu", action="store_true")
-parser.set_defaults(plot=True, gpu=False)
+parser.set_defaults(plot=True, gpu=True)
 
 args = parser.parse_args()
 
@@ -90,7 +90,7 @@
 
 # Determines number of workers to use
 if n_workers == -1:
-    n_workers = gpu * 4 * torch.cuda.device_count()
+    n_workers = 0  # gpu * 1 * torch.cuda.device_count()
 
 n_sqrt = int(np.ceil(np.sqrt(n_neurons)))
 start_intensity = intensity
@@ -116,7 +116,7 @@
 dataset = MNIST(
     PoissonEncoder(time=time, dt=dt),
     None,
-    root=os.path.join(ROOT_DIR, "data", "MNIST"),
+    "../../data/MNIST",
     download=True,
     transform=transforms.Compose(
         [transforms.ToTensor(), transforms.Lambda(lambda x: x * intensity)]

examples/mnist/conv_mnist.py

@@ -41,7 +41,7 @@
 parser.add_argument("--test", dest="train", action="store_false")
 parser.add_argument("--plot", dest="plot", action="store_true")
 parser.add_argument("--gpu", dest="gpu", action="store_true")
-parser.set_defaults(plot=True, gpu=False, train=True)
+parser.set_defaults(plot=True, gpu=True, train=True)
 
 args = parser.parse_args()
 
@@ -164,7 +164,7 @@
         start = t()
 
     train_dataloader = torch.utils.data.DataLoader(
-        train_dataset, batch_size=1, shuffle=True, num_workers=4, pin_memory=gpu
+        train_dataset, batch_size=1, shuffle=True, num_workers=0, pin_memory=gpu
     )
 
     for step, batch in enumerate(tqdm(train_dataloader)):

examples/mnist/eth_mnist.py

@@ -48,7 +48,7 @@
 parser.add_argument("--test", dest="train", action="store_false")
 parser.add_argument("--plot", dest="plot", action="store_true")
 parser.add_argument("--gpu", dest="gpu", action="store_true")
-parser.set_defaults(plot=True, gpu=False)
+parser.set_defaults(plot=True, gpu=True)
 
 args = parser.parse_args()
 
@@ -85,7 +85,7 @@
 
 # Determines number of workers to use
 if n_workers == -1:
-    n_workers = gpu * 4 * torch.cuda.device_count()
+    n_workers = 0  # gpu * 4 * torch.cuda.device_count()
 
 if not train:
     update_interval = n_test

examples/mnist/supervised_mnist.py

@@ -46,7 +46,7 @@
 parser.add_argument("--plot", dest="plot", action="store_true")
 parser.add_argument("--gpu", dest="gpu", action="store_true")
 parser.add_argument("--device_id", type=int, default=0)
-parser.set_defaults(plot=True, gpu=False, train=True)
+parser.set_defaults(plot=True, gpu=True, train=True)
 
 args = parser.parse_args()
 

requirements.txt

@@ -2,8 +2,8 @@ foolbox
 scipy>=1.1.0
 numpy>=1.14.2
 cython>=0.28.5
-torch==1.8.1
-torchvision==0.9.1
+torch>=1.5.1
+torchvision>=0.6.1
 tqdm>=4.19.9
 setuptools>=39.0.1
 matplotlib>=2.1.0

setup.py

@@ -5,7 +5,7 @@
 
 setup(
     name="BindsNET",
-    version="0.2.8",
+    version="0.2.9",
     description="Spiking neural networks for ML in Python",
     license="AGPL-3.0",
     long_description=long_description,
@@ -17,8 +17,8 @@
     zip_safe=False,
     install_requires=[
         "numpy>=1.14.2",
-        "torch==1.8.1",
-        "torchvision==0.9.1",
+        "torch>=1.5.1",
+        "torchvision>=0.6.1",
         "tensorboardX>=1.7",
         "tqdm>=4.19.9",
         "matplotlib>=2.1.0",