Merge pull request #2 from michabirklbauer/develop

add wf and tests

Author: michabirklbauer

.github/workflows/python-app.yml

@@ -0,0 +1,41 @@
+# This workflow will install Python dependencies, run tests and lint with a variety of Python versions
+# Reference workflow provided by (c) GitHub
+# For more information see: https://help.github.com/actions/language-and-framework-guides/using-python-with-github-actions
+
+name: neuralnet
+
+on:
+  push:
+    branches: [ master ]
+  pull_request:
+    branches: [ master ]
+
+jobs:
+  build:
+
+    runs-on: ubuntu-latest
+    strategy:
+      matrix:
+        python-version: ['3.7', '3.8', '3.9', '3.10']
+
+    steps:
+    - uses: actions/checkout@v2
+    - name: Set up Python ${{ matrix.python-version }}
+      uses: actions/setup-python@v2
+      with:
+        python-version: ${{ matrix.python-version }}
+    - name: Install dependencies
+      run: |
+        python -m pip install --upgrade pip
+        python -m pip install flake8 pytest
+        if [ -f requirements.txt ]; then pip install -r requirements.txt; fi
+    - name: Lint with flake8
+      run: |
+        # stop the build if there are Python syntax errors or undefined names
+        flake8 . --count --select=E9,F63,F7,F82 --show-source --statistics
+        # exit-zero treats all errors as warnings. The GitHub editor is 127 chars wide
+        flake8 . --count --exit-zero --max-complexity=10 --max-line-length=127 --statistics
+    - name: Test with pytest
+      run: |
+        cp neuralnet.py tests/neuralnet.py
+        pytest tests/tests.py

neuralnet-colab.ipynb

@@ -402,7 +402,7 @@
     "            elif self.layers[i][\"activation\"] == \"softmax\":\n",
     "                self.layers[i][\"A\"] = ActivationFunctions.softmax(self.layers[i][\"L\"])\n",
     "            else:\n",
-    "                raise NotImplementedError(\"Activation function '\" + layer[\"activation\"] + \"' not implemented!\")\n",
+    "                raise NotImplementedError(\"Activation function '\" + self.layers[i][\"activation\"] + \"' not implemented!\")\n",
     "\n",
     "    # back propagation\n",
     "    def __back_propagation(self, data: np.array, target: np.array, learning_rate: float = 0.1) -> float:\n",
@@ -494,7 +494,7 @@
     "            self.layers[-1][\"W\"] -= learning_rate * dW\n",
     "            self.layers[-1][\"b\"] -= learning_rate * db\n",
     "        else:\n",
-    "            raise NotImplementedError(\"The combination of '\" + self.loss + \" loss' and '\" + self.layers[i][\"activation\"] + \" activation' is not implemented!\")\n",
+    "            raise NotImplementedError(\"The combination of '\" + self.loss + \" loss' and '\" + self.layers[-1][\"activation\"] + \" activation' is not implemented!\")\n",
     "\n",
     "        # back propagation through the remaining hidden layers\n",
     "        for i in reversed(range(len(self.layers) - 1)):\n",

neuralnet.py

@@ -358,7 +358,7 @@ def __forward_propagation(self, data: np.array) -> None:
             elif self.layers[i]["activation"] == "softmax":
                 self.layers[i]["A"] = ActivationFunctions.softmax(self.layers[i]["L"])
             else:
-                raise NotImplementedError("Activation function '" + layer["activation"] + "' not implemented!")
+                raise NotImplementedError("Activation function '" + self.layers[i]["activation"] + "' not implemented!")
 
     # back propagation
     def __back_propagation(self, data: np.array, target: np.array, learning_rate: float = 0.1) -> float:
@@ -450,7 +450,7 @@ class labels of the input data
             self.layers[-1]["W"] -= learning_rate * dW
             self.layers[-1]["b"] -= learning_rate * db
         else:
-            raise NotImplementedError("The combination of '" + self.loss + " loss' and '" + self.layers[i]["activation"] + " activation' is not implemented!")
+            raise NotImplementedError("The combination of '" + self.loss + " loss' and '" + self.layers[-1]["activation"] + " activation' is not implemented!")
 
         # back propagation through the remaining hidden layers
         for i in reversed(range(len(self.layers) - 1)):

tests/tests.py

@@ -0,0 +1,99 @@
+#!/usr/bin/env python3
+
+# NEURAL NETWORK IMPLEMENTATION - TESTS
+# 2022 (c) Micha Johannes Birklbauer
+# https://github.com/michabirklbauer/
+# [email protected]
+
+def test_bcc():
+
+    #### Binary-class Classification ####
+
+    from zipfile import ZipFile as zip
+
+    with zip("data.zip") as f:
+        f.extractall()
+        f.close()
+
+    from neuralnet import NeuralNetwork
+    import numpy as np
+    import pandas as pd
+    from sklearn.metrics import accuracy_score
+    from sklearn.preprocessing import OneHotEncoder
+    from sklearn.model_selection import train_test_split
+
+    data = pd.read_csv("binaryclass_train.csv", header = None)
+    data["label"] = data[1].apply(lambda x: 1 if x == "M" else 0)
+    train, test = train_test_split(data, test_size = 0.3)
+    train_data = train.loc[:, ~train.columns.isin([0, 1, "label"])].to_numpy()
+    train_target = train["label"].to_numpy()
+    test_data = test.loc[:, ~test.columns.isin([0, 1, "label"])].to_numpy()
+    test_target = test["label"].to_numpy()
+
+    NN = NeuralNetwork(input_size = train_data.shape[1])
+    NN.add_layer(16, "relu")
+    NN.add_layer(16, "relu")
+    NN.add_layer(1, "sigmoid")
+    NN.compile(loss = "binary crossentropy")
+    NN.summary()
+
+    hist = NN.fit(train_data, train_target, epochs = 1000, batch_size = 32, learning_rate = 0.01)
+
+    train_predictions = np.round(NN.predict(train_data))
+    train_acc = accuracy_score(train["label"].to_numpy(), train_predictions)
+    test_predictions = np.round(NN.predict(test_data))
+    test_acc = accuracy_score(test["label"].to_numpy(), test_predictions)
+
+    import os
+    os.remove("binaryclass_train.csv")
+    os.remove("multiclass_train.csv")
+
+    assert train_acc > 0.85 and test_acc > 0.85
+
+def test_mcc():
+
+    #### Multi-class Classification ####
+
+    from zipfile import ZipFile as zip
+
+    with zip("data.zip") as f:
+        f.extractall()
+        f.close()
+
+    from neuralnet import NeuralNetwork
+    import numpy as np
+    import pandas as pd
+    from sklearn.metrics import accuracy_score
+    from sklearn.preprocessing import OneHotEncoder
+    from sklearn.model_selection import train_test_split
+
+    data = pd.read_csv("multiclass_train.csv")
+    train, test = train_test_split(data, test_size = 0.3)
+    train_data = train.loc[:, train.columns != "label"].to_numpy() / 255
+    train_target = train["label"].to_numpy()
+    test_data = test.loc[:, test.columns != "label"].to_numpy() / 255
+    test_target = test["label"].to_numpy()
+
+    one_hot = OneHotEncoder(sparse = False, categories = "auto")
+    train_target = one_hot.fit_transform(train_target.reshape(-1, 1))
+    test_target = one_hot.transform(test_target.reshape(-1, 1))
+
+    NN = NeuralNetwork(input_size = train_data.shape[1])
+    NN.add_layer(32, "relu")
+    NN.add_layer(16, "relu")
+    NN.add_layer(10, "softmax")
+    NN.compile(loss = "categorical crossentropy")
+    NN.summary()
+
+    hist = NN.fit(train_data, train_target, epochs = 30, batch_size = 16, learning_rate = 0.05)
+
+    train_predictions = np.argmax(NN.predict(train_data), axis = 1)
+    train_acc = accuracy_score(train["label"].to_numpy(), train_predictions)
+    test_predictions = np.argmax(NN.predict(test_data), axis = 1)
+    test_acc = accuracy_score(test["label"].to_numpy(), test_predictions)
+
+    import os
+    os.remove("binaryclass_train.csv")
+    os.remove("multiclass_train.csv")
+
+    assert train_acc > 0.85 and test_acc > 0.85