push main

session_61_assignment_7.11/README.md

@@ -0,0 +1,91 @@
+### Without Dropout layer
+<table>
+    <tr>
+        <td>Featue Vector Dimension</td>
+        <td>Train Loss</td>
+        <td>Train Accuracy</td>
+        <td>Test Loss</td>
+        <td>Test Accuracy</td>
+        <td>Inference Time</td>
+    </tr>
+    <tr>
+        <td>50d</td>
+        <td>0.3673</td>
+        <td>0.9394</td>
+        <td>0.4503</td>
+        <td>0.8571</td>
+        <td>0.0686s</td>
+    </tr>    
+    </tr>
+        <td>100d</td>
+        <td>...</td>
+        <td>...</td>
+        <td>...</td>
+        <td>...</td>
+        <td>...s</td>
+    </tr>
+    <tr>
+        <td>200d</td>
+        <td>...</td>
+        <td>...</td>
+        <td>...</td>
+        <td>...</td>
+        <td>...s</td>
+    </tr>
+    <tr>
+        <td>300d</td>
+        <td>...</td>
+        <td>...</td>
+        <td>...</td>
+        <td>...</td>
+        <td>...s</td>
+    </tr>
+
+</table>
+
+<br>
+
+### With Dropout layer
+<table>
+    <tr>
+        <td>Featue Vector Dimension</td>
+        <td>Train Loss</td>
+        <td>Train Accuracy</td>
+        <td>Test Loss</td>
+        <td>Test Accuracy</td>
+        <td>Inference Time</td>
+    </tr>
+    <tr>
+        <td>50d</td>
+        <td>...</td>
+        <td>...</td>
+        <td>...</td>
+        <td>...</td>
+        <td>...s</td>
+    </tr>    
+    </tr>
+        <td>100d</td>
+        <td>...</td>
+        <td>...</td>
+        <td>...</td>
+        <td>...</td>
+        <td>...s</td>
+    </tr>
+    <tr>
+        <td>200d</td>
+        <td>...</td>
+        <td>...</td>
+        <td>...</td>
+        <td>...</td>
+        <td>...s</td>
+    </tr>
+    <tr>
+        <td>300d</td>
+        <td>...</td>
+        <td>...</td>
+        <td>...</td>
+        <td>...</td>
+        <td>...s</td>
+    </tr>
+
+</table>

session_61_assignment_7.11/test.py

@@ -0,0 +1,32 @@
+from time import time
+from argparse import ArgumentParser, BooleanOptionalAction
+
+from text_classifier import EmojiTextClassifier
+from tqdm import tqdm
+
+args = ArgumentParser()
+args.add_argument('--model', type=str,
+                  required=True, help='The model path')
+args.add_argument('--vectors-file', type=str,
+                  default='./glove_6B/glove.6B.200d.txt', help='The feature vectors file path')
+args.add_argument('--sentence', type=str,
+                  required=True, help='The sentence to test the model')
+args.add_argument('--infer', type=bool, default=True,
+                  action=BooleanOptionalAction, help='Whether to inferences the model with your sentence or not')
+args.add_argument('--n-infer', type=int,
+                  default=100, help='Number of inferences on your sentence')
+
+opt = args.parse_args()
+classifier = EmojiTextClassifier(...)
+classifier.load_feature_vectors(opt.vectors_file)
+classifier.load_model(opt.model)
+emoji = classifier.predict(opt.sentence)
+print(emoji)
+
+if opt.infer:
+    start_time = time()
+    for i in tqdm(range(opt.n_infer)):
+        classifier.predict(opt.sentence)
+
+    duration = time() - start_time
+    print(f'\nAverage inference time: {duration / opt.n_infer}')

session_61_assignment_7.11/text_classifier.py

@@ -0,0 +1,113 @@
+from keras.models import Sequential, load_model
+from keras import layers
+from keras.activations import softmax
+from keras.losses import categorical_crossentropy
+from keras.optimizers import Adam
+from keras.callbacks import History, ModelCheckpoint
+
+import numpy as np
+import pandas as pd
+
+
+class EmojiTextClassifier:
+    def __init__(self, dimension: int) -> None:
+        self.dimension = dimension
+        self.model = None
+        self.words_vectors: dict[str, np.ndarray] = {}
+
+    @staticmethod
+    def load_dataset(dataset_path: str) -> list[np.ndarray, np.ndarray]:
+        df = pd.read_csv(dataset_path)
+        x = np.array(df['sentence'])
+        y = np.array(df['label'], dtype=int)
+
+        return x, y
+
+    def load_feature_vectors(self, file_path: str) -> None:
+        line: str
+
+        file = open(file_path, encoding='utf-8')
+        for line in file:
+            line = line.strip().split(' ')
+            word = line[0]
+            vector = np.array(line[1:], dtype=np.float64)
+            self.words_vectors[word] = vector
+
+    def sentence_to_feature_vectors_avg(self, sentence: str) -> np.ndarray[np.floating] | None:
+        sentence = sentence.lower()
+        words = sentence.strip().split(' ')
+        sum_vectors = np.zeros((50, ))
+
+        try:
+            for word in words:
+                vector = self.words_vectors[word]
+                sum_vectors += vector
+            return sum_vectors / len(words)
+        except KeyError:
+            print(f'There is an unknown word in this sentence: "{sentence}"')
+
+    def convert_sentences_to_vectors(self, sentences: np.ndarray) -> np.ndarray:
+        sentences_avg = []
+        for sentence in sentences:
+            sentences_avg.append(
+                self.sentence_to_feature_vectors_avg(sentence)
+            )
+
+        return np.array(sentences_avg)
+
+    def build_model(self, with_dropout: bool = False) -> None:
+        if with_dropout:
+            self.model = Sequential([
+                layers.Dropout(.5, name='DropoutLayer'),
+                layers.Dense(5, activation=softmax,
+                             input_shape=(self.dimension, ), name='OutputLayer')
+            ])
+
+        else:
+            self.model = Sequential([
+                layers.Dense(5, activation=softmax,
+                             input_shape=(self.dimension, ), name='OutputLayer')
+            ])
+
+    def compile_model(self, optimizer=Adam(), loss=categorical_crossentropy) -> None:
+        self.model.compile(
+            optimizer=optimizer,
+            loss=loss,
+            metrics=['accuracy']
+        )
+
+    def train_model(self, x_train, y_train, epochs: int, model_path_to_save: str = 'best_emojis_classifier.keras') -> History:
+        check = ModelCheckpoint(model_path_to_save,
+                                monitor='accuracy', save_best_only=True)
+
+        history: History = self.model.fit(
+            x_train,
+            y_train,
+            epochs=epochs,
+            callbacks=[check]
+        )
+
+        return history
+
+    @staticmethod
+    def evaluate(model_path, x_test, y_test) -> None:
+        model: Sequential = load_model(model_path)
+        print('\nEvaluating model...')
+        model.evaluate(x_test, y_test)
+
+    def load_model(self, model_path: str) -> None:
+        self.model: Sequential = load_model(model_path)
+
+    def predict(self, sentence: str) -> str:
+        sentence_avg = self.sentence_to_feature_vectors_avg(sentence)
+        sentence_avg = np.array([sentence_avg])
+
+        prediction = self.model.predict(sentence_avg)
+        y_hat = np.argmax(prediction)   
+
+        return EmojiTextClassifier.covert_label_to_emoji(y_hat)
+
+    @staticmethod
+    def covert_label_to_emoji(label: int) -> str:
+        emojis = ['🧡', '⚾', '😃', '😔', '🍴']
+        return emojis[label]

session_61_assignment_7.11/train.py

@@ -0,0 +1,56 @@
+from argparse import ArgumentParser, BooleanOptionalAction
+
+from keras.callbacks import History
+from keras.utils import to_categorical
+import matplotlib.pyplot as plt
+from text_classifier import EmojiTextClassifier
+
+args = ArgumentParser()
+args.add_argument('--train-dataset', type=str,
+                  default='./dataset/train.csv', help='The train dataset path')
+args.add_argument('--test-dataset', type=str,
+                  default='./dataset/test.csv', help='The test dataset path')
+args.add_argument('--dimension', type=int,
+                  default=50, help='The dimension of feature vectors')
+args.add_argument('--vectors-file', type=str,
+                  default='./glove_6B/glove.6B.200d.txt', help='The feature vectors file path')
+args.add_argument('--dropout', type=bool,
+                  default=False, action=BooleanOptionalAction, help='Add dropout layer to network')
+args.add_argument('--model-save', type=str,
+                  default='best_emojis_classifier.keras', help='The best model path to save')
+args.add_argument('--epochs', type=int,
+                  default=200, help='The number of epochs to train the model')
+args.add_argument('--save-plots', type=bool,
+                  default=True, action=BooleanOptionalAction, help='Save the training information plots')
+
+opt = args.parse_args()
+
+classifier = EmojiTextClassifier(opt.dimension)
+x_train, y_train = EmojiTextClassifier.load_dataset(opt.train_dataset)
+x_test, y_test = EmojiTextClassifier.load_dataset(opt.test_dataset)
+
+classifier.load_feature_vectors(opt.vectors_file)
+x_train = classifier.convert_sentences_to_vectors(x_train)
+x_test = classifier.convert_sentences_to_vectors(x_test)
+
+y_train = to_categorical(y_train)
+y_test = to_categorical(y_test)
+
+classifier.build_model(opt.dropout)
+classifier.compile_model()
+history: History = classifier.train_model(
+    x_train, y_train, opt.epochs, opt.model_save)
+
+EmojiTextClassifier.evaluate(opt.model_save, x_test, y_test)
+
+if opt.save_plots:
+    fig, (ax1, ax2) = plt.subplots(1, 2)[1]
+    ax1.plot(history.history['accuracy'])
+    ax1.set_xlabel('Epochs')
+    ax1.set_ylabel('Accuracy')
+
+    ax2.plot(history.history['loss'])
+    ax2.set_xlabel('Epochs')
+    ax2.set_ylabel('Loss')
+    fig.suptitle(f'Emojis Classification With Dropout: {False}')
+    plt.savefig(f'Emojis_Classification_{opt.epochs}ep_dropout_{opt.dropout}.png')