Deep8 is a dynamic Deep Learning framework like Chainer and DyNet.
Change the setting in CMakeList.txt line 4:
HAVE_GPU: if build for GPU
BUILD_PYTHON: if build for python
BUILD_TEST: if build the Test
Build C++
# Set the HAVE_GPU be TRUE if have a GPU in CMakeList.txt
# Set BUILD_TEST be TRUE to build a Test
# input the follow command
# run the "deep8_test" to Test
# add the "deep8_native.so" in C++ project
mkdir build
cd build
cmake ..
makeBuild for Python
# Set the HAVE_GPU be TRUE if have a GPU in CMakeList.txt
# Set BUILD_PYTHON be TRUE
# Run below cmd in Deep8 folder
# import the "deep8" in Python project
python setup.py install# coding=utf-8
import numpy as np
from deep8 import *
executor = EagerExecutor()
learningRate = ConstantLearningRateIterator(0.01)
trainer = SGDTrainer(learningRate = learningRate)
'''
|4, -1| |a| |10|
| | * | | = | | ====> a = 3, b = 2
|2, 1| |b| |8 |
'''
x = np.array([4, -1, 2, 1], dtype=np.float32)
y = np.array([10, 8], dtype=np.float32)
w = parameter(executor, [2])
w.gaussian()
input = parameter(executor, [2, 2], False)
output = parameter(executor, [2], False)
input.feed(x)
output.feed(y)
for i in range(1000):
(input * w - output).l1NormLoss().backward()
trainer.train(executor)
print i + 1, "=>", w.valueStr()
print "The w should be around [3, 2]: ", w.valueStr()/**
* |4, -1| |a| |10|
* | | * | | = | | ====> a = 3, b = 2
* |2, 1| |b| |8 |
*/
float x[4] = { 4, -1, 2, 1 };
float y[2] = { 10, 8 };
EagerExecutor executor;
LinearDecayLearningRateIterator learningRate(1000);
AdamTrainer trainer(&learningRate);
auto w = parameter(&executor, { 2 });
w.gaussian();
auto input = parameter(&executor, { 2, 2 }, false);
auto output = parameter(&executor, { 2 }, false);
input.feed(x);
output.feed(y);
for (int i = 0; i < 1000; ++i) {
(input * w - output).l1NormLoss().backward();
trainer.train(&executor, executor.trainableParameters());
/**print the w*/
std::cout << i + 1 << " => " << w.valueStr() << std::endl;
}
std::cout << "the result should be around: [3, 2]" << std::endl;