Helper tools I made to work with tensorflow
dependecies:
tensorflow, pydub, soundfile, pickle
Examples:
- Minst classifier
from helpers.extractor import *
from helpers.neural_network import *
loader = MinstLoader("data/minst_folder/")
network = ConvNet(loader)
network.Run()- Cifar classifier
from helpers.extractor import *
from helpers.neural_network import *
loader = CifarLoader("data/cifar10", "Grayscale")
network = ConvNet(loader)
network.Run()- Customize graph:
from helpers.extractor import *
from helpers.neural_network import *
layers = []
layers.append(NNOperation("reshape", [-1, 28, 28, 1]))
layers.append(NNOperation("conv2d", [5, 5, 1, 32]))
layers.append(NNOperation("maxpool2d", 2))
layers.append(NNOperation("conv2d", [5, 5, 32, 64]))
layers.append(NNOperation("maxpool2d", 2))
layers.append(NNOperation("reshape", [-1, 7*7*64]))
layers.append(NNOperation("wx+b", [7*7*64, 1024], [1024]))
layers.append(NNOperation("relu"))
layers.append(NNOperation("dropout", 0.75))
layers.append(NNOperation("wx+b", [1024, 10], [10]))
loader = CifarLoader("data/cifar10", "Grayscale")
network = ConvNet(loader)
network.Run(layers)- Load and Save graph:
from helpers.extractor import *
from helpers.neural_network import *
loader = CifarLoader("data/cifar10", "Grayscale")
network = ConvNet(loader, save_path="graphs/Cifar10Graph", restore_path="graphs/Cifar10Graph")
network.Run(layers)- MultiLayer Dynamic LSTM Example:
from helpers.extractor import *
from helpers.neural_network import *
loader = CifarLoader("inputs/datasets/cifar10/", "RGB")
network = ConvNet(loader,
n_steps=32*3,
training_iters=100000,
display_step=1,
learning_rate = 0.001,
batch_size=128)
layers = []
layers.append(LSTMOperation(cells=[1024, 256, 10], n_classes=network.n_classes))
x=tf.placeholder("float", [None, 3*32, 32])
network.Run(x=x, layers=layers, save_path="graphs/Cifar10Graph")- Conv-LSTM graph:
from helpers.extractor import *
from helpers.neural_network import *
loader = CifarLoader("inputs/datasets/cifar10/", "RGB")
network = ConvNet(loader,
training_iters=100000,
display_step=1,
learning_rate = 0.001,
batch_size=128)
layers = []
layers.append(NNOperation("reshape", [-1, 32*3, 32, 1]))
layers.append(NNOperation("conv2d", [3, 3, 1, 32]))
layers.append(NNOperation("maxpool2d", 2)) #16
layers.append(NNOperation("conv2d", [3, 3, 32, 64]))
layers.append(NNOperation("maxpool2d", 2)) #8
layers.append(NNOperation("reshape", [-1, 24*64, 8]))
layers.append(LSTMOperation(cells=[32], n_classes=network.n_classes))
network.Run(layers=layers, save_path="graphs/Cifar10Graph")