girija2204

from sklearn.preprocessing import LabelEncoder

encoder = LabelEncoder()
encoder = encoder.fit(tps_data.cat0)
cat0_encoded_le = encoder.transform(tps_data.cat0)

pd.value_counts(cat0_encoded_le)
# 0    223525
# 1     76475
# dtype: int64

girija2204

def encode_leaveoneout(dataset,column):
    encoder = LeaveOneOutEncoder()
    encoder = encoder.fit(dataset[column],dataset['target'])
    dataset[f"{column}_loo_enc"] = encoder.transform(dataset[column])
    return dataset

def encode_label(dataset,column):
    encoder = LabelEncoder()
    encoder = encoder.fit(dataset[column])
    dataset[f"{column}_le_enc"] = encoder.transform(dataset[column])

girija2204

from category_encoders import LeaveOneOutEncoder

encoder = LeaveOneOutEncoder()
encoder = encoder.fit(tps_data.cat0,tps_data.target)
cat0_encoded_loo = encoder.transform(tps_data.cat0)

pd.value_counts(cat0_encoded_loo['cat0'])
# 0.334070    223525
# 0.062609     76475
# Name: cat0, dtype: int64

girija2204

tps_data = pd.read_csv("../input/tabular-playground-series-mar-2021/train.csv")
tps_data.columns
# Index(['id', 'cat0', 'cat1', 'cat2', 'cat3', 'cat4', 'cat5', 'cat6', 'cat7',
#        'cat8', 'cat9', 'cat10', 'cat11', 'cat12', 'cat13', 'cat14', 'cat15',
#        'cat16', 'cat17', 'cat18', 'cont0', 'cont1', 'cont2', 'cont3', 'cont4',
#        'cont5', 'cont6', 'cont7', 'cont8', 'cont9', 'cont10', 'target'],
#       dtype='object')

pd.value_counts(tps_data.cat0)
# A    223525

girija2204

images_pair, labels_pair = generate_train_image_pairs(images_dataset, labels_dataset)
history = model.fit([images_pair[:, 0], images_pair[:, 1]], labels_pair[:],validation_split=0.1,batch_size=64,epochs=100)

girija2204

image = images_dataset[92] # a random image as test image
test_image_pairs, test_label_pairs = generate_test_image_pairs(images_dataset, labels_dataset, image) # produce an array of test image pairs and test label pairs

# for each pair in the test image pair, predict the similarity between the images
for index, pair in enumerate(test_image_pairs):
    pair_image1 = np.expand_dims(pair[0], axis=-1)
    pair_image1 = np.expand_dims(pair_image1, axis=0)
    pair_image2 = np.expand_dims(pair[1], axis=-1)
    pair_image2 = np.expand_dims(pair_image2, axis=0)
    prediction = model.predict([pair_image1, pair_image2])[0][0]

girija2204

def generate_test_image_pairs(images_dataset, labels_dataset, image):
    unique_labels = np.unique(labels_dataset)
      label_wise_indices = dict()
      for label in unique_labels:
          label_wise_indices.setdefault(label,
                                        [index for index, curr_label in enumerate(labels_dataset) if
                                         label == curr_label])
  
    pair_images = []
    pair_labels = []

girija2204

def generate_train_image_pairs(images_dataset, labels_dataset):
    unique_labels = np.unique(labels_dataset)
    label_wise_indices = dict()
    for label in unique_labels:
        label_wise_indices.setdefault(label,
                                      [index for index, curr_label in enumerate(labels_dataset) if
                                       label == curr_label])
    
    pair_images = []
    pair_labels = []

girija2204

model.compile(loss="binary_crossentropy", optimizer="adam", metrics=["accuracy"])

girija2204

def euclidean_distance(vectors):
    (featA, featB) = vectors
    sum_squared = k.sum(k.square(featA - featB), axis=1, keepdims=True)
    return k.sqrt(k.maximum(sum_squared, k.epsilon()))

distance = Lambda(euclidean_distance)([featA, featB])
outputs = Dense(1, activation="sigmoid")(distance)
model = Model(inputs=[imgA, imgB], outputs=outputs)