Classification

Procfile

@@ -0,0 +1 @@
+worker: python classification_app.py

classification/lenet.py

@@ -1,4 +1,3 @@
-from abc import ABC
 from typing import Iterable
 
 import keras

classification/neural_network.py

@@ -1,16 +1,10 @@
-import json
-from datetime import datetime
-from typing import Iterable, Union, Optional
+from typing import Iterable, Union
 from pathlib import Path
 from operator import itemgetter
 from abc import ABC, abstractmethod
 
-from keras.wrappers.scikit_learn import KerasClassifier
+
 from tensorflow.keras import Model
-from tensorflow.keras.utils import to_categorical
-from sklearn.model_selection import train_test_split
-from sklearn.metrics import classification_report, cohen_kappa_score, confusion_matrix
-import matplotlib.pyplot as plt
 import numpy as np
 
 from preprocessing.image_edition import image_to_square, resize_threshold
@@ -38,52 +32,20 @@ def model(self) -> Model:
     def labels(self) -> Iterable[str]:
         pass
 
-    def get_train_test_sets(self, *, data_file: Union[str, Path]):
-        database = np.load(data_file)
-        dataset, labels = database['database'], database['label']
-        labels = to_categorical(labels)
-        # split dataset in train and test
-        x_train, x_test, y_train, y_test = train_test_split(dataset, labels, test_size=0.1, stratify=labels)
-        x_train = x_train.reshape(x_train.shape[0], *self.input_shape)
-        x_test = x_test.reshape(x_test.shape[0], *self.input_shape)
-        self.train_set = (x_train, y_train)
-        self.test_set = (x_test, y_test)
-
-    def save_weights(self, file_path: Union[str, Path], *, save_format: str = 'h5'):
-        self.model.save_weights(file_path, save_format=save_format)
-
     def load_weights(self, file: Union[str, Path]):
         self.model.load_weights(file)
 
-    def train(self, *, data_file: Union[str, Path], batch_size: int = 32, epochs: int = 10,
-              logs_dir: Optional[Union[str, Path]] = None, weights_dir=None):
-        # load data
-        self.get_train_test_sets(data_file=data_file)
-        history = self.model.fit(*self.train_set, epochs=epochs, batch_size=batch_size, validation_data=self.test_set)
-        self.history = history.history
-        if logs_dir or weights_dir:
-            now = datetime.now()
-            date_string = now.strftime('%Y_%m_%d_%H_%M_%S')
-            data_string = f'd={date_string}_e={epochs}_b={batch_size}'
-        if logs_dir:
-            with open(f'{logs_dir}/loss_history_{self.__class__.__name__.lower()}_{data_string}.log', 'w') as file:
-                file.write(json.dumps(self.history, indent=4, sort_keys=True))
-            accuracy, cohen_kappa = self.accuracy_metrics()
-            with open(f'{logs_dir}/accuracy_metrics_{self.__class__.__name__.lower()}_{data_string}.log', 'w') as file:
-                file.write(accuracy)
-                file.write(f'Cohen Kappa score : {cohen_kappa}')
-        if weights_dir:
-            self.save_weights(f'{weights_dir}/{self.__class__.__name__.lower()}_weights_{data_string}.h5',
-                              save_format='h5')
-
     def predict(self, img: np.ndarray):
         return self.model.predict(img)
 
     def predict_array(self, array: Iterable[np.ndarray]):
         predictions_results = [None] * len(array)
         for img_idx, img in enumerate(array):
-
-            squared_img = image_to_square(img)
+            try:
+                squared_img = image_to_square(img)
+            except ValueError:
+                predictions_results[img_idx] = ""
+                continue
             resized_img = resize_threshold(squared_img, self.input_shape[:2])
             input_img = resized_img[np.newaxis, :, :, np.newaxis]
             prediction_array = self.predict(input_img)
@@ -94,39 +56,3 @@ def predict_array(self, array: Iterable[np.ndarray]):
             prediction = sorted(prediction_list, key=itemgetter(1))[-1]
             predictions_results[img_idx] = prediction[0]
         return predictions_results
-
-    def accuracy_metrics(self):
-        y_pred = self.model.predict(self.test_set[0], batch_size=32, verbose=1)
-        y_pred_max = np.argmax(y_pred, axis=1)
-
-        y_true = []
-        for i in range(len(self.test_set[1])):
-            y_true.append(np.argmax(self.test_set[1][i]))
-
-        accuracy = classification_report(y_true, y_pred_max)
-        cohen_kappa = cohen_kappa_score(y_true, y_pred_max)
-        conf_matrix = confusion_matrix(y_true, y_pred_max)
-        return accuracy, cohen_kappa, conf_matrix
-
-    def train_test_plt(self):
-        fig = plt.figure(figsize=(15, 5))
-        fig.add_subplot(1, 2, 1)
-
-        plt.plot(self.history['val_loss'])
-        plt.plot(self.history['loss'])
-        plt.title('Model Loss')
-        plt.xlabel('Epoch')
-        plt.ylabel('Loss')
-        plt.legend(['Test', 'Train'], loc='upper left')
-
-        fig.add_subplot(1, 2, 2)
-        plt.plot(self.history['val_accuracy'])
-        plt.plot(self.history['accuracy'])
-        plt.title('Model Accuracy')
-        plt.xlabel('Epoch')
-        plt.ylabel('Accuracy')
-        plt.legend(['Test', 'Train'], loc='upper left')
-
-        fig.subplots_adjust(wspace=0.2, hspace=1)
-
-

dataset/label_dict_2021_02_27_01_08_49.txt

@@ -15,41 +15,41 @@
   "13": "8",
   "14": "9",
   "15": "a",
-  "16": "alpha",
+  "16": "\\alpha",
   "17": "b",
-  "18": "beta",
+  "18": "\\beta",
   "19": "c",
-  "20": "cos",
+  "20": "\\cos",
   "21": "d",
-  "22": "delta",
+  "22": "\\delta",
   "23": "e",
   "24": "f",
-  "25": "forward_slash",
+  "25": "/",
   "26": "g",
-  "27": "gamma",
+  "27": "\\gamma",
   "28": "h",
   "29": "i",
   "30": "j",
   "31": "k",
   "32": "l",
-  "33": "lambda",
-  "34": "log",
+  "33": "\\lambda",
+  "34": "\\log",
   "35": "m",
-  "36": "mu",
+  "36": "\\mu",
   "37": "n",
   "38": "o",
   "39": "p",
-  "40": "phi",
-  "41": "pi",
+  "40": "\\phi",
+  "41": "\\pi",
   "42": "q",
   "43": "r",
   "44": "s",
-  "45": "sigma",
-  "46": "sin",
-  "47": "sqrt",
+  "45": "\\sigma",
+  "46": "\\sin",
+  "47": "\\sqrt",
   "48": "t",
-  "49": "tan",
-  "50": "theta",
+  "49": "\\tan",
+  "50": "\\theta",
   "51": "u",
   "52": "v",
   "53": "w",