fixed svm for not posdef kernel matrix, updated .travis.yml with Python 3.8 and added svr with r2 and accuracy metrics (aimacode#1185)

Author: dmeoli

.travis.yml

@@ -4,27 +4,13 @@ python:
   - 3.5
   - 3.6
   - 3.7
+  - 3.8
 
 before_install:
   - git submodule update --remote
 
 install:
-  - pip install flake8
-  - pip install ipython
-  - pip install ipythonblocks
-  - pip install ipywidgets
-  - pip install keras
-  - pip install matplotlib
-  - pip install networkx
-  - pip install numpy
-  - pip install opencv-python
-  - pip install Pillow
-  - pip install pytest-cov
-  - pip install qpsolvers
-  - pip install quadprog
-  - pip install six
-  - pip install sortedcontainers
-  - pip install tensorflow
+  - pip install --upgrade -r requirements.txt
 
 script:
   - py.test --cov=./

csp.py

@@ -758,8 +758,9 @@ class Sudoku(CSP):
     . . 2 | 6 . 9 | 5 . .
     8 . . | 2 . 3 | . . 9
     . . 5 | . 1 . | 3 . .
-    >>> AC3(e); e.display(e.infer_assignment())
-    (True, 6925)
+    >>> AC3(e)  # doctest: +ELLIPSIS
+    (True, ...)
+    >>> e.display(e.infer_assignment())
     4 8 3 | 9 2 1 | 6 5 7
     9 6 7 | 3 4 5 | 8 2 1
     2 5 1 | 8 7 6 | 4 9 3
@@ -1265,7 +1266,7 @@ def display(self, assignment=None):
                 else:
                     var = "p" + str(j) + str(i)
                     if assignment is not None:
-                        if isinstance(assignment[var], set) and len(assignment[var]) is 1:
+                        if isinstance(assignment[var], set) and len(assignment[var]) == 1:
                             puzzle += "[" + str(first(assignment[var])).upper() + "] "
                         elif isinstance(assignment[var], str):
                             puzzle += "[" + str(assignment[var]).upper() + "] "
@@ -1393,7 +1394,7 @@ def display(self, assignment=None):
                         var2 = "0" + var2
                     var = "X" + var1 + var2
                     if assignment is not None:
-                        if isinstance(assignment[var], set) and len(assignment[var]) is 1:
+                        if isinstance(assignment[var], set) and len(assignment[var]) == 1:
                             puzzle += "[" + str(first(assignment[var])) + "]\t"
                         elif isinstance(assignment[var], int):
                             puzzle += "[" + str(assignment[var]) + "]\t"

deep_learning4e.py

@@ -8,7 +8,7 @@
 from keras.layers import Embedding, SimpleRNN, Dense
 from keras.preprocessing import sequence
 
-from utils4e import (softmax1D, conv1D, gaussian_kernel, element_wise_product, vector_add, random_weights,
+from utils4e import (conv1D, gaussian_kernel, element_wise_product, vector_add, random_weights,
                      scalar_vector_product, map_vector, mean_squared_error_loss)
 
 
@@ -46,6 +46,9 @@ def function(self, x):
     def derivative(self, x):
         return NotImplementedError
 
+    def __call__(self, x):
+        return self.function(x)
+
 
 class Sigmoid(Activation):
 
@@ -56,7 +59,7 @@ def derivative(self, value):
         return value * (1 - value)
 
 
-class Relu(Activation):
+class ReLU(Activation):
 
     def function(self, x):
         return max(0, x)
@@ -65,13 +68,28 @@ def derivative(self, value):
         return 1 if value > 0 else 0
 
 
-class Elu(Activation):
+class ELU(Activation):
+
+    def __init__(self, alpha=0.01):
+        self.alpha = alpha
 
-    def function(self, x, alpha=0.01):
-        return x if x > 0 else alpha * (np.exp(x) - 1)
+    def function(self, x):
+        return x if x > 0 else self.alpha * (np.exp(x) - 1)
 
-    def derivative(self, value, alpha=0.01):
-        return 1 if value > 0 else alpha * np.exp(value)
+    def derivative(self, value):
+        return 1 if value > 0 else self.alpha * np.exp(value)
+
+
+class LeakyReLU(Activation):
+
+    def __init__(self, alpha=0.01):
+        self.alpha = alpha
+
+    def function(self, x):
+        return max(x, self.alpha * x)
+
+    def derivative(self, value):
+        return 1 if value > 0 else self.alpha
 
 
 class Tanh(Activation):
@@ -83,13 +101,31 @@ def derivative(self, value):
         return 1 - (value ** 2)
 
 
-class LeakyRelu(Activation):
+class SoftMax(Activation):
+
+    def function(self, x):
+        return np.exp(x) / np.sum(np.exp(x))
+
+    def derivative(self, x):
+        return np.ones_like(x)
+
+
+class SoftPlus(Activation):
 
-    def function(self, x, alpha=0.01):
-        return x if x > 0 else alpha * x
+    def function(self, x):
+        return np.log(1. + np.exp(x))
+
+    def derivative(self, x):
+        return 1. / (1. + np.exp(-x))
 
-    def derivative(self, value, alpha=0.01):
-        return 1 if value > 0 else alpha
+
+class Linear(Activation):
+
+    def function(self, x):
+        return x
+
+    def derivative(self, x):
+        return np.ones_like(x)
 
 
 class InputLayer(Layer):
@@ -112,9 +148,9 @@ class OutputLayer(Layer):
     def __init__(self, size=3):
         super().__init__(size)
 
-    def forward(self, inputs):
+    def forward(self, inputs, activation=SoftMax):
         assert len(self.nodes) == len(inputs)
-        res = softmax1D(inputs)
+        res = activation().function(inputs)
         for node, val in zip(self.nodes, res):
             node.value = val
         return res