tfkeras_convpredict.py

from __future__ import print_function
import tensorflow as tf
import tensorflow.keras as keras
from tensorflow.keras.models import Sequential
from tensorflow.keras.layers import Dense, Dropout, Flatten
from tensorflow.keras.layers import Conv2D, MaxPooling2D
import tensorflow_addons as tfa
import os
import numpy as np
import tamil
import copy
import glob
from matplotlib import pyplot as plt
from time import sleep
# Adapted from: https://keras.io/examples/mnist_cnn/

# 1) Setup letters to be built
uyir_plus_ayutham = copy.copy(tamil.utf8.uyir_letters)
uyir_plus_ayutham.append( tamil.utf8.ayudha_letter )

def to_tamil_letter(idx):
    return(uyir_plus_ayutham[idx])

def load_acchu_data(mode='train'):
    path = os.path.split(__file__)[0]
    labels_path = os.path.join(path,'data',mode+'-label-onehot.npy')
    images_path = os.path.join(path,'data',mode+'-image.npy')
    labels = np.load(labels_path)
    images = np.load(images_path)
    return labels,images

batch_size = 128
num_classes = 13
epochs = 200

# input image dimensions
img_rows, img_cols = 28, 28

# the data, split between train and test sets
y_train, x_train  = load_acchu_data('train')
y_test, x_test    = load_acchu_data('test')
#y_test=y_test[:100,:]; x_test=x_test[:100,:];


x_train = x_train.reshape(len(x_train), img_rows, img_cols)
x_test = x_test.reshape(len(x_test), img_rows, img_cols)
input_shape = (img_rows, img_cols)

x_train = x_train.astype('float32')
x_test = x_test.astype('float32')
x_train /= 255.0
x_test /= 255.0
print('x_train shape:', x_train.shape)
print(x_train.shape[0], 'train samples')
print(x_test.shape[0], 'test samples')

#model = keras.models.load_model('acchu_conv_model_5')
#model = keras.models.load_model('acchu_conv3_model_2')
model = keras.models.load_model('acchu_model_4')
model.summary()

if False:
    figure = plt.figure(figsize=(8, 8))
    sns.heatmap(con_mat_df, annot=True,cmap=plt.cm.Blues)
    plt.tight_layout()
    plt.ylabel('True label')
    plt.xlabel('Predicted label')
    plt.show()

y_pred = model.predict_classes(x_test)
y_test_index = np.zeros(y_pred.shape[0])
y_pred_index = np.zeros(y_pred.shape[0])
for i in range(y_pred.shape[0]):
    y_test_index[i]=np.argmax(y_test)
    y_pred_index[i]=np.argmax(y_pred)
print(y_test_index)
print(y_pred_index)
con_mat = tf.math.confusion_matrix(labels=y_test_index, predictions=y_pred_index).numpy()
print(con_mat)
raise Exception()

expected = ['a','aa','e','ee','u','uu','eh','aeh','ai','o','oh','au','ak']
assert len(expected) == 13
stats = 0.0
total = 0.0
msg = []
outputF1 = tfa.metrics.F1Score(num_classes=num_classes,average='macro')
for letter_image in glob.glob('letters-hand-drawn-corrected/*.npy'):#npy/
    total += 1.0
    print("#"*32)
    data = np.load(letter_image)
    #data = data.transpose()
    plt.imshow(data)
    plt.show()
    #sleep(5)
    data = data.reshape(1,28,28,1)/255.0
    output = model.predict(data)
    print(letter_image)
    predicted = np.argmax(output[0])
    print("predicted class=>",to_tamil_letter(predicted))
    sfx = letter_image.split('_')[1].replace('.npy','')
    print(sfx)
    actuals = np.zeros((1,13))
    outputP = np.zeros((1,13))
    actuals[0,expected.index(sfx)]=1.0
    outputP[0,predicted]=1.0
    print(actuals)
    print(output[0])
    outputF1.update_state(actuals.astype(np.float32), outputP.astype(np.float32))
    if predicted != expected.index(sfx):
        stats += 1.0
        msg.append("{0} misclassified as {1}".format(to_tamil_letter(expected.index(sfx)),to_tamil_letter(predicted)))
    print(output[0])
print("Failed cases: %g = (%g/%g)"%(stats/total,stats,total))
print("\n".join(msg))
print('F1 Macro score is: ',outputF1.result().numpy())
if False:
    score = model.evaluate(x_test, y_test, verbose=1)
    print('Test loss:', score[0])
    print('Test accuracy:', score[1])