hypercube-cascade 1.0.2__tar.gz → 1.0.3__tar.gz

{hypercube_cascade-1.0.2 → hypercube_cascade-1.0.3}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: hypercube-cascade
-Version: 1.0.2
+Version: 1.0.3
 Summary: Python bindings for HypercubeCascade: frozen etalon transit + frozen reservoir orbit + HypercubeCNN on end state
 License-Expression: Apache-2.0
 Classifier: Development Status :: 5 - Production/Stable
@@ -199,14 +199,21 @@ techniques on spectra like these, and very likely beats them.
 
 ![Held-out validation extract, spectra 581 through 584](https://raw.githubusercontent.com/dliptak001/HypercubeCascade/main/examples/RamanBaselineExtraction/extracted_baselines_cascade.png)
 
-### Etalon sets the bar. Cascade raises it (well, maybe).
+### Three hosts, one floor
 
 The etalon-only sibling
 ([HypercubeEtalon](https://github.com/dliptak001/HypercubeEtalon))
-is this Cascade with the reservoir removed, and on its own it
-already does everything described above. With the very same
-Exciter and readout configuration, its overlays are
-indistinguishable from the ones shown.
+is this Cascade with the reservoir removed. The reservoir-only
+sibling ([HypercubeWTF](https://github.com/dliptak001/HypercubeWTF))
+is this Cascade with the transit removed: the same frozen reservoir
+— same seed, spectral radius, history depth, and pass count — driven
+by the normalized spectrum directly. Each of them, on its own,
+already does everything described above. With the very same readout
+configuration and the same 60-epoch budget, the Etalon scores 4.77
+and WTF 4.76 against the Cascade's 4.82, and all three overlays are
+indistinguishable from the one shown. Three preprocessors that share
+no mechanism — a transit, an orbit, and the two in series — carry
+the same one-layer, one-channel readout to the same floor.
 
 Real spectra, however, are not nearly this clean. Low laser power,
 short integration times, and weak scatterers all put noise on the
@@ -217,11 +224,16 @@ That is what the Cascade's second stage, the Reservoir, is for. On
 the strength of the MNIST white-noise study
 ([`examples/mnist/WhiteNoiseFilter.md`](https://github.com/dliptak001/HypercubeCascade/blob/main/examples/mnist/WhiteNoiseFilter.md)),
 the Cascade is expected to outperform the Etalon alone in that
-noise.
+noise — and WTF's own study
+([`examples/mnist/WhiteNoiseFilter.md`](https://github.com/dliptak001/HypercubeWTF/blob/main/examples/mnist/WhiteNoiseFilter.md))
+found that same reservoir a filter that holds accuracy as the noise
+rises. Whether the transit in front of it adds anything under noise,
+or whether the reservoir is doing all of the filtering, is the open
+question.
 
 That is the next experiment.
 
-Side-by-side overlays and both training profiles are in
+Side-by-side overlays and all three training profiles are in
 [`examples/RamanBaselineExtraction/`](https://github.com/dliptak001/HypercubeCascade/blob/main/examples/RamanBaselineExtraction/README.md).
 
 ---

{hypercube_cascade-1.0.2 → hypercube_cascade-1.0.3}/README.md

@@ -172,14 +172,21 @@ techniques on spectra like these, and very likely beats them.
 
 ![Held-out validation extract, spectra 581 through 584](https://raw.githubusercontent.com/dliptak001/HypercubeCascade/main/examples/RamanBaselineExtraction/extracted_baselines_cascade.png)
 
-### Etalon sets the bar. Cascade raises it (well, maybe).
+### Three hosts, one floor
 
 The etalon-only sibling
 ([HypercubeEtalon](https://github.com/dliptak001/HypercubeEtalon))
-is this Cascade with the reservoir removed, and on its own it
-already does everything described above. With the very same
-Exciter and readout configuration, its overlays are
-indistinguishable from the ones shown.
+is this Cascade with the reservoir removed. The reservoir-only
+sibling ([HypercubeWTF](https://github.com/dliptak001/HypercubeWTF))
+is this Cascade with the transit removed: the same frozen reservoir
+— same seed, spectral radius, history depth, and pass count — driven
+by the normalized spectrum directly. Each of them, on its own,
+already does everything described above. With the very same readout
+configuration and the same 60-epoch budget, the Etalon scores 4.77
+and WTF 4.76 against the Cascade's 4.82, and all three overlays are
+indistinguishable from the one shown. Three preprocessors that share
+no mechanism — a transit, an orbit, and the two in series — carry
+the same one-layer, one-channel readout to the same floor.
 
 Real spectra, however, are not nearly this clean. Low laser power,
 short integration times, and weak scatterers all put noise on the
@@ -190,11 +197,16 @@ That is what the Cascade's second stage, the Reservoir, is for. On
 the strength of the MNIST white-noise study
 ([`examples/mnist/WhiteNoiseFilter.md`](https://github.com/dliptak001/HypercubeCascade/blob/main/examples/mnist/WhiteNoiseFilter.md)),
 the Cascade is expected to outperform the Etalon alone in that
-noise.
+noise — and WTF's own study
+([`examples/mnist/WhiteNoiseFilter.md`](https://github.com/dliptak001/HypercubeWTF/blob/main/examples/mnist/WhiteNoiseFilter.md))
+found that same reservoir a filter that holds accuracy as the noise
+rises. Whether the transit in front of it adds anything under noise,
+or whether the reservoir is doing all of the filtering, is the open
+question.
 
 That is the next experiment.
 
-Side-by-side overlays and both training profiles are in
+Side-by-side overlays and all three training profiles are in
 [`examples/RamanBaselineExtraction/`](https://github.com/dliptak001/HypercubeCascade/blob/main/examples/RamanBaselineExtraction/README.md).
 
 ---

{hypercube_cascade-1.0.2 → hypercube_cascade-1.0.3}/hypercube_cascade/_version.py

@@ -2,4 +2,4 @@
 # - pyproject.toml reads this via scikit-build-core dynamic metadata
 # - hypercube_cascade.__version__ imports it
 # - bindings.cpp gets the same string at compile time via CMake
-__version__ = "1.0.2"
+__version__ = "1.0.3"