Improved confusion matrix generation, so it uses samples that are roughly...

Improved confusion matrix generation, so it uses samples that are roughly uniformly distributed between the 2 classes, across the entire validation set. The matrix agrees with the reported accuracy, suggesting something is wrong with my implementation used with OpenCV in c++ (as the acc. is lower).

output/Output CSV/Output 2021-07-11.csv

-Detection_ID,Date,Time of day (hour:min),Source_file,Time in (min:sek),Time out (min:sek),Art
-0,2021-07-11,17:03:37,Myggbukta-[2021-07-09_13-48-44]-105,00:09,00:18,
-1,2021-07-11,17:03:37,Myggbukta-[2021-07-09_13-48-44]-105,00:40,00:43,
-2,2021-07-11,17:03:37,Myggbukta-[2021-07-09_13-48-44]-105,01:28,02:30,
-3,2021-07-11,17:03:37,Myggbukta-[2021-07-09_13-48-44]-105,04:20,21:05,
-4,2021-07-11,17:03:37,Myggbukta-[2021-07-09_13-48-44]-105,21:42,21:57,
-5,2021-07-11,17:03:37,Myggbukta-[2021-07-09_13-48-44]-105,21:59,22:02,
-0,2021-07-11,17:03:37,Myggbukta-[2021-07-09_13-48-44]-105,00:09,00:18,
-1,2021-07-11,17:03:37,Myggbukta-[2021-07-09_13-48-44]-105,00:40,00:43,
-2,2021-07-11,17:03:37,Myggbukta-[2021-07-09_13-48-44]-105,01:28,02:30,
-3,2021-07-11,17:03:37,Myggbukta-[2021-07-09_13-48-44]-105,04:20,21:05,
-4,2021-07-11,17:03:37,Myggbukta-[2021-07-09_13-48-44]-105,21:42,21:57,
-5,2021-07-11,17:03:37,Myggbukta-[2021-07-09_13-48-44]-105,21:59,22:02,
-0,2021-07-11,17:03:37,Myggbukta-[2021-07-09_13-48-44]-105,00:09,00:18,
-1,2021-07-11,17:03:37,Myggbukta-[2021-07-09_13-48-44]-105,00:40,00:43,
-2,2021-07-11,17:03:37,Myggbukta-[2021-07-09_13-48-44]-105,01:28,02:30,
-3,2021-07-11,17:03:37,Myggbukta-[2021-07-09_13-48-44]-105,04:20,21:05,
-4,2021-07-11,17:03:37,Myggbukta-[2021-07-09_13-48-44]-105,21:42,21:57,
-5,2021-07-11,17:03:37,Myggbukta-[2021-07-09_13-48-44]-105,21:59,22:02,

src/python/utilities/visualizeTrainHist.py

-
-from tkinter import E
-
-import matplotlib.pyplot as plt
-import json
-
-def plot_graphs(history, string):
-    plt.plot(history[string])
-    plt.plot(history["val_"+string])
-    plt.xlabel("Epochs")
-    plt.ylabel(string)
-    plt.legend([string,"val_"+string])
-    plt.show()
-
-if __name__ == "__main__":
-    history = json.load(open("src/python/utilities/trainedModels/3/trainingHistory.json"))
-    for key in history:
-        print(key)
-    plot_graphs(history,'accuracy')
-    plot_graphs(history,'loss')
\ No newline at end of file

src/python/utilities/visualizeTrainResults.py

+
+import matplotlib.pyplot as plt
+import seaborn as sn
+import json
+import numpy as np
+import tensorflow as tf
+from tensorflow.keras.preprocessing import image_dataset_from_directory
+from tensorflow.keras import Model
+from tensorflow.keras import models
+
+def plot_graphs(history, string):
+    plt.plot(history[string])
+    plt.plot(history["val_"+string])
+    plt.xlabel("Epochs")
+    plt.ylabel(string)
+    plt.legend([string,"val_"+string])
+    plt.show()
+
+def plot_confusionMatrix():
+    resizeWidth = 288
+    resizeHeight = 162
+
+    datasetLimit = 1500
+
+    model = models.load_model("src\\python\\utilities\\trainedModels\\3\\model")
+    dataset = image_dataset_from_directory(
+        directory="dataset/validate",
+        labels="inferred",
+        label_mode="binary",
+        class_names=["NoFish", "fish"],
+        image_size=(resizeWidth, resizeHeight),
+        batch_size=1,
+        shuffle=False,
+	    interpolation="nearest"
+    )
+
+    labelsNumpy = []
+    data = []
+    iterator = dataset.as_numpy_iterator()
+    step = int(dataset.cardinality().numpy()/datasetLimit)
+    
+    for cout in range(datasetLimit):
+        img, label = iterator.next()
+        labelsNumpy.append(label[0])
+        data.append(img)
+        for i in range(step-1):
+            iterator.next()
+    data = np.array(data)
+    data = np.squeeze(data)
+    
+    labelsNumpy = np.array(labelsNumpy)
+    labelsNumpy.flatten()
+
+    preds = model.predict(data, batch_size=1)
+    preds.flatten()
+
+    #   2-by-2 matrix for binary classification.
+    matrix = np.zeros((2,2), dtype=np.int32)
+    for i in range(len(labelsNumpy)):
+        y = int(round(preds[i][0], 1))
+        x = int(round(labelsNumpy[i][0], 1))
+        matrix[y][x] += 1
+
+    classes = ["Not fish", "Fish"]
+    sn.set(font_scale=2)
+
+    ax = sn.heatmap(matrix, annot=True, xticklabels=classes, yticklabels=classes, cbar=True, fmt="g")
+    ax.set(title="Confusion matrix", xlabel="Ground truth", ylabel="Predicted")
+    plt.tight_layout()
+    plt.show()
+
+
+if __name__ == "__main__":
+    history = json.load(open("src/python/utilities/trainedModels/3/trainingHistory.json"))
+    plot_graphs(history,'accuracy')
+    plot_graphs(history,'loss')
+    plot_confusionMatrix()