FFNN-IRIS: complete example of ffnn using rich

Author: ludekcizinsky

examples/ffnn-iris/.gitignore

@@ -0,0 +1,3 @@
+# Ignore produced outputs
+*.html
+*.png

examples/ffnn-iris/main.py

@@ -5,41 +5,54 @@
 from nnlearn.network import FFNN, DenseLayer
 from nnlearn.datasets import load_iris
 from nnlearn.metrics import accuracy_score
+from nnlearn.util import ScriptInformation
 
 # TODO: replace this with own implentation
 from sklearn import preprocessing
 from sklearn.model_selection import train_test_split
 
 
 def test_ffnn_classifier():
-
-    # Load and split data
+    
+    logger = ScriptInformation()
+    logger.section_start(":herb: FFNN - IRIS")
+    logger.script_time()
+    logger.author("Ludek", "Cizinsky")
+    logger.section_start(":construction: Prepare input for the model")
+
+    logger.working_on("Load and split data") 
     X, y = load_iris()
 
-    # Process the data
+    logger.working_on("Process the data")
     y = preprocessing.LabelEncoder().fit_transform(y)
     X = preprocessing.StandardScaler().fit_transform(X)
 
-    # Train test split
+    logger.working_on("Train test split")
     X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.33, random_state=42)
     m = X_train.shape[1]
-
-    # Define the layers
+ 
+    logger.working_on("Define the layers")
     layers = [DenseLayer(m, 3), DenseLayer(3, 4), DenseLayer(4, 3, activation='softmax')]
 
-    # Train the model 
+    logger.section_start(":robot: Train the model")
     clf = FFNN(layers,
-               epochs = 1,
+               logger=logger,
+               epochs = 50,
                loss_func='cross_entropy',
                batch_size=.5,
                lr=.9,
                shuffle=True)
     clf.fit(X_train, y_train)
+    
+    clf.fig.savefig("training.png", bbox_inches='tight')
+    logger.c.print(clf.report)
 
-    # Validate the model
+    logger.section_start(":crystal_ball: Validate the model")
     y_hat = clf.predict(X_test)
-    print(f'Validation accuracy: {accuracy_score(y_test, y_hat)}')
+    acc = accuracy_score(y_test, y_hat)
+    logger.important_metric('Accuracy', acc)
 
+    logger.save("ffnn-iris.html")
 
 if __name__ == '__main__':
     test_ffnn_classifier()

examples/template/main.py

@@ -2,9 +2,11 @@
 import sys
 sys.path.insert(0, os.path.abspath('../../'))
 
-def your_test():
-    print("Hello world!")
+from nnlearn.util import ScriptInformation
 
+def your_test():
+    s = ScriptInformation()
+    s.hello()
 
 if __name__ == '__main__':
     your_test()

nnlearn/network/_base.py

@@ -2,20 +2,31 @@
 import numpy as np
 
 class Base:
+    """
+    Base for all parametric models which are optimized via Gradient descent.
+    """
 
     def __init__(self):
       pass
 
     @staticmethod
     def _arr_to_var(X):
-
         """
         Transform given X values into Var objects.
 
         Notes
         -----
         Assumption made: X is an nd-numpy array with numerical variables.
         """
-            
-        return np.vectorize(lambda x: Var(x))(X)
+        
+        f = np.vectorize(lambda x: Var(x))
+        return f(X)
+    
+    @staticmethod
+    def _arr_to_val(X):
+        """
+        Get from given X's Var instances their values.
+        """
+        f = np.vectorize(lambda x: x.v)
+        return f(X)