adjust structure

split models package, add tools package

dalib/__init__.py

@@ -1,2 +1 @@
-from .models import *
-from .datasets import *
+__all__ = ['datasets', 'adaptation', 'vision', 'text']

dalib/adaptation/__init__.py

@@ -0,0 +1 @@
+__all__ = ["cdan", "dann", "mdd", "dan", "afn"]

dalib/models/afn.py → dalib/adaptation/afn.py

@@ -1,6 +1,6 @@
 import math
 import torch.nn as nn
-from .classifier import Classifier as ClassifierBase
+from dalib.vision.classifier import Classifier as ClassifierBase
 
 
 class StepwiseAdaptiveFeatureNorm(nn.Module):
@@ -30,6 +30,21 @@ def forward(self, features):
         return ((features.norm(p=2, dim=1) - radius) ** 2).mean()
 
 
+class L2PreservedDropout(nn.Module):
+    """Dropout that preserves the L 1-norm in both of the training and evaluation phases.
+    .. TODO math definitions
+    """
+    def __init__(self, *args, **kwargs):
+        super(L2PreservedDropout, self).__init__()
+        self.dropout = nn.Dropout(*args, **kwargs)
+
+    def forward(self, input):
+        output = self.dropout(input)
+        if self.training:
+            output.mul_(math.sqrt(1 - self.dropout.p))
+        return output
+
+
 class Classifier(ClassifierBase):
     def __init__(self, backbone, num_classes, bottleneck_dim=1000, dropout_p=0.5):
         bottleneck = nn.Sequential(
@@ -48,17 +63,3 @@ def get_parameters(self):
         ]
         return params
 
-
-class L2PreservedDropout(nn.Module):
-    """Dropout that preserves the L 1-norm in both of the training and evaluation phases.
-    .. TODO math definitions
-    """
-    def __init__(self, *args, **kwargs):
-        super(L2PreservedDropout, self).__init__()
-        self.dropout = nn.Dropout(*args, **kwargs)
-
-    def forward(self, input):
-        output = self.dropout(input)
-        if self.training:
-            output.mul_(math.sqrt(1 - self.dropout.p))
-        return output

dalib/models/cdan.py → dalib/adaptation/cdan.py

@@ -3,14 +3,14 @@
 import torch.nn as nn
 import torch.nn.functional as F
 from ._util import binary_accuracy, WarmStartGradientReverseLayer
-from .classifier import Classifier as ClassifierBase
+from dalib.vision.classifier import Classifier as ClassifierBase
 
 
 __all__ = ['DomainDiscriminator', 'ConditionalDomainAdversarialLoss']
 
 
 class DomainDiscriminator(nn.Module):
-    """Domain discriminator model. See class:`dalib.models.dann.DomainDiscriminator` for details.
+    """Domain discriminator model. See class:`dalib.adaptation.dann.DomainDiscriminator` for details.
     """
     def __init__(self, in_feature, hidden_size):
         super(DomainDiscriminator, self).__init__()

dalib/models/dan.py → dalib/adaptation/dan.py

@@ -1,6 +1,6 @@
 import torch
 import torch.nn as nn
-from .classifier import Classifier as ClassifierBase
+from dalib.vision.classifier import Classifier as ClassifierBase
 
 __all__ = ['MultipleKernelMaximumMeanDiscrepancy', 'GaussianKernel']
 
@@ -22,7 +22,7 @@ class MultipleKernelMaximumMeanDiscrepancy(nn.Module):
         >>> feature_dim = 1024
         >>> batch_size = 10
         >>> kernels = [GaussianKernel(alpha=0.5), GaussianKernel(1.), GaussianKernel(2.)]
-        >>> loss = models.dan.MultipleKernelMaximumMeanDiscrepancy(kernels)
+        >>> loss = adaptation.dan.MultipleKernelMaximumMeanDiscrepancy(kernels)
         >>> # features from source domain and target domain
         >>> f_s, f_t = torch.randn(batch_size, feature_dim), torch.randn(batch_size, feature_dim)
         >>> output = loss(f_s, f_t)

dalib/models/dann.py → dalib/adaptation/dann.py

@@ -1,7 +1,7 @@
 import torch
 import torch.nn as nn
 from ._util import WarmStartGradientReverseLayer, binary_accuracy
-from .classifier import Classifier as ClassifierBase
+from dalib.vision.classifier import Classifier as ClassifierBase
 
 __all__ = ['DomainDiscriminator', 'DomainAdversarialLoss']
 

dalib/models/mdd.py → dalib/adaptation/mdd.py

@@ -24,7 +24,7 @@ class MarginDisparityDiscrepancyLoss(nn.Module):
 
     .. note::
         MarginDisparityDiscrepancyLoss has already used GradientReverseLayer, thus adversarial_classifier is
-        merely a classifier head, e.g. `dalib.models.mdd.AdversarialClassifier`
+        merely a classifier head, e.g. `dalib.adaptation.mdd.AdversarialClassifier`
 
     Shape:
         - y_s, y_t: :math:`(N, C)` where C = number of classes.
@@ -127,7 +127,7 @@ def forward(self, x):
 
 
 class Classifier(nn.Module):
-    """Classifier for MDD. Similar as `nn.dalib.models.classifier.Classifier`"""
+    """Classifier for MDD. Similar as `nn.dalib.vision.classifier.Classifier`"""
     def __init__(self, backbone, num_classes, use_bottleneck=True, bottleneck_dim=1024, head_bottleneck_dim=1024):
         super(Classifier, self).__init__()
         self.backbone = backbone

dalib/vision/__init__.py

@@ -0,0 +1,3 @@
+from .resnet import *
+
+__all__ = ['resnet', 'classifier']

docs/Makefile

@@ -1,10 +1,10 @@
 # Minimal makefile for Sphinx documentation
 #
 
-# You can set these variables from the command line.
-SPHINXOPTS    =
-SPHINXBUILD   = sphinx-build
-SPHINXPROJ    = dalib
+# You can set these variables from the command line, and also
+# from the environment for the first two.
+SPHINXOPTS    ?=
+SPHINXBUILD   ?= sphinx-build
 SOURCEDIR     = source
 BUILDDIR      = build
 
@@ -17,4 +17,4 @@ help:
 # Catch-all target: route all unknown targets to Sphinx using the new
 # "make mode" option.  $(O) is meant as a shortcut for $(SPHINXOPTS).
 %: Makefile
-	@$(SPHINXBUILD) -M $@ "$(SOURCEDIR)" "$(BUILDDIR)" $(SPHINXOPTS) $(O)
+	@$(SPHINXBUILD) -M $@ "$(SOURCEDIR)" "$(BUILDDIR)" $(SPHINXOPTS) $(O)

docs/source/dalib.adaptation.rst

@@ -0,0 +1,55 @@
+dalib.adaptation package
+========================
+
+Module contents
+---------------
+
+.. automodule:: dalib.adaptation
+
+dalib.adaptation.dan module
+---------------------------
+
+.. autoclass:: dalib.adaptation.dan.MultipleKernelMaximumMeanDiscrepancy
+    :show-inheritance:
+
+.. autoclass:: dalib.adaptation.dan.GaussianKernel
+    :show-inheritance:
+
+dalib.adaptation.dann module
+----------------------------
+
+.. autoclass:: dalib.adaptation.dann.DomainAdversarialLoss
+    :show-inheritance:
+
+.. autoclass:: dalib.adaptation.dann.DomainDiscriminator
+    :show-inheritance:
+
+dalib.adaptation.cdan module
+----------------------------
+
+.. autoclass:: dalib.adaptation.cdan.ConditionalDomainAdversarialLoss
+    :show-inheritance:
+
+.. autoclass:: dalib.adaptation.cdan.DomainDiscriminator
+    :show-inheritance:
+
+dalib.adaptation.mdd module
+----------------------------
+
+.. autoclass:: dalib.adaptation.mdd.MarginDisparityDiscrepancyLoss
+    :show-inheritance:
+
+.. autoclass:: dalib.adaptation.mdd.MarginDisparityDiscrepancy
+    :show-inheritance:
+
+.. autoclass:: dalib.adaptation.mdd.AdversarialClassifier
+    :show-inheritance:
+
+.. autoclass:: dalib.adaptation.mdd.Classifier
+    :show-inheritance:
+
+dalib.adaptation.afn module
+----------------------------
+
+.. autoclass:: dalib.adaptation.afn.StepwiseAdaptiveFeatureNorm
+    :show-inheritance:

docs/source/dalib.rst

@@ -7,7 +7,8 @@ Subpackages
 .. toctree::
 
     dalib.datasets
-    dalib.models
+    dalib.adaptation
+    dalib.vision
 
 Module contents
 ---------------

docs/source/dalib.text.rst

@@ -0,0 +1,10 @@
+dalib.text package
+==================
+
+Module contents
+---------------
+
+.. automodule:: dalib.text
+   :members:
+   :undoc-members:
+   :show-inheritance:

docs/source/dalib.vision.rst

@@ -0,0 +1,25 @@
+dalib.vision package
+====================
+
+Module contents
+---------------
+
+.. automodule:: dalib.vision
+   :members:
+   :show-inheritance:
+
+dalib.vision.classifier module
+------------------------------
+
+.. autoclass:: dalib.vision.classifier.Classifier
+    :show-inheritance:
+
+
+dalib.vision.resnet module
+--------------------------
+
+.. automodule:: dalib.vision.resnet
+   :members:
+   :undoc-members:
+   :show-inheritance:
+