Press the flag. The characters are sorted by how similarly they're used. For example, see how the vowels are in a little cluster. You occasionally have to press the flag twice #word2vec #word #embedding #machine #learning # neural #network #nlp #natural #language #processing
This uses one-hot encoding and a neural network. Trained 100 percent in Scratch. See inside for the neural network code. How does this work? It runs through all the training data, and for each letter, uses one-hot encoding for the 33 neurons of the input layer, aiming for the output layer to output the encoding of the letter that comes after it. Then, after training, it inputs the encoding of each letter again and sets the x and y to neurons one and two of the middle layer. 33-neuron input layer, linear two-neuron middle layer, softmax 33-neuron output layer
