This is the documentation for "Optical Character Recognition with Machine Learning"
This collection is made for Python Interpreter version 3.8
The Required Python Libraries to run the algorithms are:
-matplotlib -keras -tensorflow -opencv-python -scikit-learn -numpy
DATASET Features 1-) SKLearn Digits Dataset = 1797 (88 array) (0-15 white/black level) 2-) MNist Digits Dataset = 70000 (2828 array) (0-255 white/black level)
ALGORITHM Features
-
KNN n_neighbors = 3 / Number of neighbors weights = uniform / {uniform, distance} algorithm = auto / {auto, ball_tree, kd_tree, brute}
-
SVM kernel = linear / {linear, poly, rbf, sigmoid, precomputed} degree = 3 / Degree of the polynomial kernel gamma = scale / {scale, auto} Kernel coefficient
-
NN layers = 3 / {linear, poly, rbf, sigmoid, precomputed} optimizer = adam / {SGD,RMSprop, Adam, Adadelta, Adagrad, Adamax, Nadam, Ftrl} epochs = 5 / default=3 batch_size = default=32 / 1875*32 = 60.000 image (for Mnist) class_weight = default=None sample_weight = default=None loss = sparse_categorical_crossentropy
-
RFC sample_weight = None / Must be equal size weight array as samples n_estimators = default=100 / The number of trees in the forest. max_depth = default=None / The maximum depth of the tree. decision_function_shape =default='ovr / {'ovo', 'ovr'} one-vs-one, one-vs-rest