Merge pull request #20 from LukasHedegaard/develop

Bug-fixes in forward_stepping and clean_state

CHANGELOG.md

@@ -1,24 +1,32 @@
 # Changelog
 All notable changes to this project will be documented in this file.
 
-The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.0.0/),
-and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
+The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.0.0/).
+
+This project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html), with the exception that v0.X updates include backwards-incompatible API changes.
+From v1.0.0 and on, the project will adherence strictly to Semantic Versioning.
 
 
 ## [Unreleased]
 
 
+## [0.8.1]
+### Fixed
+- Bug in `forward_stepping`.
+- Bug in `clean_state`.
+
+
 ## [0.8.0]
 ### Fixed
-- Bugs in `forward_step(s)` with `update_state=False`
+- Bugs in `forward_step(s)` with `update_state=False`.
 
 ### Changed
 - `forward_steps` interface to always include `pad_end` argument.
-- name of "interface.py" to "module.py".
-- implementations of `forward_step(s)` to be consolidated in CoModule.
+- Name of "interface.py" to "module.py".
+- Implementations of `forward_step(s)` to be consolidated in CoModule.
 
 ### Removed
-- `Padded` interface
+- `Padded` interface.
 
 
 ## [0.7.0]

continual/conv.py

@@ -121,9 +121,12 @@ def init_state(
         return state_buffer, state_index, stride_index
 
     def clean_state(self):
-        del self.state_buffer
-        del self.state_index
-        del self.stride_index
+        if hasattr(self, "state_buffer"):
+            del self.state_buffer
+        if hasattr(self, "state_index"):
+            del self.state_index
+        if hasattr(self, "stride_index"):
+            del self.stride_index
 
     def get_state(self):
         if (

continual/convert.py

@@ -59,14 +59,21 @@ def call(x: Tensor, pad_end=False, update_state=True) -> Tensor:
 
         return call
 
-    def clean_state(*args, **kwargs):
+    def dummy(*args, **kwargs):
         ...  # pragma: no cover
 
+    @staticmethod
+    def build_from(mod):  # pragma: no cover
+        return module.__class__()
+
     module.forward = module.forward
     module.forward_steps = forward_steps(module.forward)
     module.forward_step = forward_step(module.forward)
     module.delay = 0
-    module.clean_state = clean_state
+    module.get_state = dummy
+    module.set_state = dummy
+    module.clean_state = dummy
+    module.build_from = build_from
 
     return module
 

continual/delay.py

@@ -44,8 +44,10 @@ def init_state(
         return state_buffer, state_index
 
     def clean_state(self):
-        del self.state_buffer
-        del self.state_index
+        if hasattr(self, "state_buffer"):
+            del self.state_buffer
+        if hasattr(self, "state_index"):
+            del self.state_index
 
     def get_state(self):
         if (

continual/pooling.py

@@ -158,9 +158,12 @@ def init_state(
         return state_buffer, state_index, stride_index
 
     def clean_state(self):
-        del self.state_buffer
-        del self.state_index
-        del self.stride_index
+        if hasattr(self, "state_buffer"):
+            del self.state_buffer
+        if hasattr(self, "state_index"):
+            del self.state_index
+        if hasattr(self, "stride_index"):
+            del self.stride_index
 
     def get_state(self):
         if (

setup.py

@@ -25,7 +25,7 @@ def from_file(file_name: str = "requirements.txt", comment_char: str = "#"):
 
 setup(
     name="continual-inference",
-    version="0.8.0",
+    version="0.8.1",
     description="Building blocks for Continual Inference Networks in PyTorch",
     long_description=long_description(),
     long_description_content_type="text/markdown",

tests/continual/test_conv.py

@@ -454,6 +454,7 @@ def test_update_state_false():
         padding_mode="zeros",
     )
     coconv = co.Conv3d.build_from(conv)
+    coconv.clean_state()  # Nothing should happen
 
     target = conv.forward(sample)
 

tests/continual/test_delay.py

@@ -9,6 +9,7 @@
 def test_delay_3d():
     sample = torch.normal(mean=torch.zeros(4 * 3 * 3)).reshape((1, 1, 4, 3, 3))
     delay = Delay(delay=2, temporal_fill="zeros")
+    delay.clean_state()  # Nothing should happen
 
     ones = torch.ones_like(sample[:, :, 0])
 

tests/continual/test_pool.py

@@ -40,6 +40,7 @@ def test_AvgPool1d_padded():
     target = pool(sample)
 
     co_pool = AvgPool1d.build_from(pool)
+    co_pool.clean_state()  # Nothing should happen
 
     # forward
     output2 = co_pool.forward(sample)