# %% [code] # This Python 3 environment comes with many helpful analytics libraries installed # It is defined by the kaggle/python Docker image: https://github.com/kaggle/docker-python # For example, here's several helpful packages to load from matplotlib import pyplot as plt import numpy as np # linear algebra import pandas as pd # data processing, CSV file I/O (e.g. pd.read_csv) import PIL.Image as Image, PIL.ImageDraw as ImageDraw, PIL.ImageFont as ImageFont import random import os import cv2 import gc from tqdm.auto import tqdm import numpy as np import keras from tensorflow.keras.models import Sequential from tensorflow.keras.layers import Dense, Dropout, Flatten from tensorflow.keras.layers import Conv2D, MaxPooling2D from tensorflow.keras.optimizers import SGD from tensorflow.keras.layers import BatchNormalization from tensorflow.keras.layers import LeakyReLU from tensorflow.keras.models import clone_model from sklearn.model_selection import train_test_split from tensorflow.keras.callbacks import ReduceLROnPlateau, EarlyStopping, ModelCheckpoint from tensorflow.keras.preprocessing.image import ImageDataGenerator from tensorflow.keras.utils import plot_model import tensorflow as tf from tensorflow import keras from tensorflow.keras import layers import os for dirname, _, filenames in os.walk('/kaggle/input'): for filename in filenames: print(os.path.join(dirname, filename)) # You can write up to 5GB to the current directory (/kaggle/working/) that gets preserved as output when you create a version using "Save & Run All" # You can also write temporary files to /kaggle/temp/, but they won't be saved outside of the current session # %% [code] def resize(df, size=64, need_progress_bar=True): resized = {} for i in range(df.shape[0]): image = cv2.resize(df.loc[df.index[i]].values.reshape(137,236),(size,size)) resized[df.index[i]] = image.reshape(-1) resized = pd.DataFrame(resized).T return resized # %% [code] train_data = pd.read_csv('/kaggle/input/bengaliai-cv19/train.csv') train_data = pd.merge(pd.read_parquet(f'/kaggle/input/bengaliai-cv19/train_image_data_0.parquet'), train_data, on='image_id').drop(['image_id'], axis=1) #train_data_1 = pd.merge(pd.read_parquet(f'/kaggle/input/bengaliai-cv19/train_image_data_1.parquet'), train_data, on='image_id').drop(['image_id'], axis=1) #train_data_2 = pd.merge(pd.read_parquet(f'/kaggle/input/bengaliai-cv19/train_image_data_2.parquet'), train_data, on='image_id').drop(['image_id'], axis=1) #train_data_3 = pd.merge(pd.read_parquet(f'/kaggle/input/bengaliai-cv19/train_image_data_3.parquet'), train_data, on='image_id').drop(['image_id'], axis=1) # %% [code] #training_data=train_data_0 #training_data=training_data.append(train_data_1,ignore_index=True) #training_data=training_data.append(train_data_2,ignore_index=True) #training_data.shape # %% [code] train_labels = train_data[['grapheme_root', 'vowel_diacritic', 'consonant_diacritic','grapheme']] train_data = train_data.drop(['grapheme_root', 'vowel_diacritic', 'consonant_diacritic','grapheme'], axis=1) train_data = resize(train_data)/255 train_data = train_data.values.reshape(-1, 64, 64, 1) # %% [code] model_dict = { 'grapheme_root': Sequential(), 'vowel_diacritic': Sequential(), 'consonant_diacritic': Sequential() } # %% [code] for model_type, model in model_dict.items(): model.add(Conv2D(32, 5, activation="relu", padding="same", input_shape=[64, 64, 1])) model.add(layers.BatchNormalization(momentum=0.15)) model.add(MaxPooling2D(2)) model.add(Conv2D(64, 3, activation="relu", padding="same")) model.add(Conv2D(64, 3, activation="relu", padding="same")) model.add(MaxPooling2D(2)) model.add(Conv2D(128, 3, activation="relu", padding="same")) model.add(Conv2D(128, 3, activation="relu", padding="same")) model.add(MaxPooling2D(2)) model.add(Conv2D(256, 3, activation="relu", padding="same")) model.add(Conv2D(256, 3, activation="relu", padding="same")) model.add(MaxPooling2D(2)) model.add(Flatten()) model.add(Dense(1024, activation="relu")) model.add(Dropout(0.5)) model.add(Dense(512, activation="relu")) model.add(Dropout(0.5)) if model_type == 'grapheme_root': model.add(layers.Dense(168, activation='softmax', name='root_out')) elif model_type == 'vowel_diacritic': model.add(layers.Dense(11, activation='softmax', name='vowel_out')) elif model_type == 'consonant_diacritic': model.add(layers.Dense(7, activation='softmax', name='consonant_out')) model.compile(optimizer="adam", loss=['categorical_crossentropy'], metrics=['accuracy']) # %% [code] plot_model(model_dict['grapheme_root']) # %% [code] batch_size = 32 epochs = 10 history_list = [] # %% [code] keras.backend.clear_session() np.random.seed(42) tf.random.set_seed(42) model_types = ['grapheme_root', 'vowel_diacritic', 'consonant_diacritic'] for target in model_types: Y_train = train_labels[target] Y_train = pd.get_dummies(Y_train).values x_train, x_test, y_train, y_test = train_test_split(train_data, Y_train, test_size=0.1, random_state=123) y_train = tf.cast(y_train, tf.int32) y_test = tf.cast(y_test, tf.int32) # datagen = ImageDataGenerator(rotation_range=5) datagen = ImageDataGenerator() checkpoint_cb = ModelCheckpoint("{}.h5".format(target), save_best_only=True) early_stopping_cb = EarlyStopping(patience=3, restore_best_weights=True) history = model_dict[target].fit_generator(datagen.flow(x_train, y_train, batch_size=batch_size), epochs = epochs, validation_data = (x_test, y_test), callbacks=[checkpoint_cb, early_stopping_cb]) history_list.append(history) model_dict[target] = keras.models.load_model("{}.h5".format(target)) del Y_train del x_train del x_test del y_train del y_test gc.collect() # %% [code] plt.figure() for i in range(3): plt.plot(np.arange(0, len(history_list[i].history['accuracy'])), history_list[i].history['accuracy'], label='train_accuracy') plt.plot(np.arange(0, len(history_list[i].history['accuracy'])), history_list[i].history['val_accuracy'], label='val_accuracy') plt.title(model_types[i]) plt.xlabel('Epoch') plt.ylabel('Accuracy') plt.legend(loc='lower right') plt.show() # %% [code] preds_dict = { 'grapheme_root': [], 'vowel_diacritic': [], 'consonant_diacritic': [] } # %% [code] target=[] # model predictions placeholder row_id=[] # row_id place holder for i in range(4): print("Parquet: {}".format(i)) df_test_img = pd.read_parquet('/kaggle/input/bengaliai-cv19/test_image_data_{}.parquet'.format(i)) df_test_img.set_index('image_id', inplace=True) X_test = resize(df_test_img, need_progress_bar=False)/255 X_test = X_test.values.reshape(-1, 64, 64, 1) for i, p in preds_dict.items(): model = keras.models.load_model("{}.h5".format(i)) preds = model.predict(X_test) preds_dict[i] = np.argmax(preds, axis=1) for k,id in enumerate(df_test_img.index.values): for i,comp in enumerate(model_types): id_sample=id+'_'+comp row_id.append(id_sample) target.append(preds_dict[comp][k]) del df_test_img del X_test gc.collect() df_sample = pd.DataFrame( { 'row_id': row_id, 'target':target }, columns = ['row_id','target'] ) df_sample.to_csv('submission.csv',index=False) # %% [code] # %% [code]