Merge ../Marian.hieu

This commit is contained in:
Hieu Hoang 2016-09-14 16:36:43 +02:00
commit 063dd8649e
6 changed files with 55 additions and 34 deletions

View File

@ -10,7 +10,7 @@ Expr::Expr(Chainable<Tensor>* chainable) : pimpl_(chainable) {}
Expr::Expr(Float v) : pimpl_(new ConstantNode(keywords::value=v, Expr::Expr(Float v) : pimpl_(new ConstantNode(keywords::value=v,
keywords::shape={1,1})) {} keywords::shape={1,1})) {}
Tensor Expr::val() { Tensor &Expr::val() {
return pimpl_->val(); return pimpl_->val();
} }

View File

@ -15,7 +15,7 @@ class Expr {
return *this; return *this;
} }
Tensor val(); Tensor &val();
Tensor grad(); Tensor grad();
void forward(size_t batchSize); void forward(size_t batchSize);

View File

@ -17,7 +17,7 @@ struct Chainable {
virtual void allocate(size_t) = 0; virtual void allocate(size_t) = 0;
virtual const Shape& shape() = 0; virtual const Shape& shape() = 0;
virtual DataType val() = 0; virtual DataType &val() = 0;
virtual DataType grad() = 0; virtual DataType grad() = 0;
virtual void setVal(Tensor t) { virtual void setVal(Tensor t) {
UTIL_THROW2("Tensors can only be assigned to input nodes"); UTIL_THROW2("Tensors can only be assigned to input nodes");
@ -82,7 +82,7 @@ class Node : public Chainable<Tensor>,
} }
} }
virtual Tensor val() { virtual Tensor &val() {
UTIL_THROW_IF2(!val_, "Tensor has not been allocated"); UTIL_THROW_IF2(!val_, "Tensor has not been allocated");
return val_; return val_;
}; };
@ -104,4 +104,4 @@ class Node : public Chainable<Tensor>,
Tensor adj_; Tensor adj_;
}; };
} }

View File

@ -80,12 +80,12 @@ void Tensor::Load(const std::string &path)
} }
strm.close(); strm.close();
Load(hostData); Load(hostData.begin(), hostData.begin());
} }
void Tensor::Load(const std::vector<float> &values) void Tensor::Load(const std::vector<float>::const_iterator &begin, const std::vector<float>::const_iterator &end)
{ {
pimpl_->set(values); pimpl_->set(begin, end);
} }
} }

View File

@ -158,11 +158,11 @@ class TensorImpl {
thrust::fill(data_.begin(), data_.end(), value); thrust::fill(data_.begin(), data_.end(), value);
} }
void set(const std::vector<Float> &values) { void set(const std::vector<float>::const_iterator &begin, const std::vector<float>::const_iterator &end) {
size_t totSize = GetTotalSize(shape()); size_t totSize = GetTotalSize(shape());
std::cerr << "tensor size=" << totSize << " vector size=" << values.size() << std::endl; //std::cerr << "tensor size=" << totSize << " vector size=" << values.size() << std::endl;
assert(totSize == values.size()); //assert(totSize == values.size());
thrust::copy(values.begin(), values.end(), data_.begin()); thrust::copy(begin, end, data_.begin());
} }
std::string Debug() const std::string Debug() const
@ -275,7 +275,7 @@ class Tensor {
} }
void Load(const std::string &path); void Load(const std::string &path);
void Load(const std::vector<float> &values); void Load(const std::vector<float>::const_iterator &begin, const std::vector<float>::const_iterator &end);
}; };

View File

@ -12,6 +12,7 @@ int main(int argc, char** argv) {
using namespace marian; using namespace marian;
using namespace keywords; using namespace keywords;
const size_t BATCH_SIZE = 500;
const size_t IMAGE_SIZE = 784; const size_t IMAGE_SIZE = 784;
const size_t LABEL_SIZE = 10; const size_t LABEL_SIZE = 10;
@ -20,42 +21,62 @@ int main(int argc, char** argv) {
Expr w = param(shape={IMAGE_SIZE, LABEL_SIZE}, name="W0"); Expr w = param(shape={IMAGE_SIZE, LABEL_SIZE}, name="W0");
Expr b = param(shape={1, LABEL_SIZE}, name="b0"); Expr b = param(shape={1, LABEL_SIZE}, name="b0");
Expr z = dot(x, w) + b; Expr z = dot(x, w) + b;
Expr lr = softmax(z, axis=1, name="pred"); Expr lr = softmax(z, axis=1, name="pred");
Expr graph = -mean(sum(y * log(lr), axis=1), axis=0, name="cost"); Expr graph = -mean(sum(y * log(lr), axis=1), axis=0, name="cost");
//cerr << "lr=" << Debug(lr.val().shape()) << endl; //cerr << "x=" << Debug(lr.val().shape()) << endl;
int numofdata; int numofdata;
vector<float> images = datasets::mnist::ReadImages("../examples/mnist/t10k-images-idx3-ubyte", numofdata, IMAGE_SIZE); //vector<float> images = datasets::mnist::ReadImages("../examples/mnist/t10k-images-idx3-ubyte", numofdata, IMAGE_SIZE);
vector<float> labels = datasets::mnist::ReadLabels("../examples/mnist/t10k-labels-idx1-ubyte", numofdata, LABEL_SIZE); //vector<float> labels = datasets::mnist::ReadLabels("../examples/mnist/t10k-labels-idx1-ubyte", numofdata, LABEL_SIZE);
vector<float> images = datasets::mnist::ReadImages("../examples/mnist/train-images-idx3-ubyte", numofdata, IMAGE_SIZE);
vector<float> labels = datasets::mnist::ReadLabels("../examples/mnist/train-labels-idx1-ubyte", numofdata, LABEL_SIZE);
cerr << "images=" << images.size() << " labels=" << labels.size() << endl; cerr << "images=" << images.size() << " labels=" << labels.size() << endl;
cerr << "numofdata=" << numofdata << endl; cerr << "numofdata=" << numofdata << endl;
Tensor tx({numofdata, IMAGE_SIZE}, 1); size_t startInd = 0;
Tensor ty({numofdata, LABEL_SIZE}, 1); size_t startIndData = 0;
while (startInd < numofdata) {
size_t batchSize = (startInd + BATCH_SIZE < numofdata) ? BATCH_SIZE : numofdata - startInd;
cerr << "startInd=" << startInd
<< " startIndData=" << startIndData
<< " batchSize=" << batchSize << endl;
tx.Load(images); Tensor tx({numofdata, IMAGE_SIZE}, 1);
ty.Load(labels); Tensor ty({numofdata, LABEL_SIZE}, 1);
cerr << "tx=" << Debug(tx.shape()) << endl; tx.Load(images.begin() + startIndData, images.begin() + startIndData + batchSize * IMAGE_SIZE);
cerr << "ty=" << Debug(ty.shape()) << endl; ty.Load(labels.begin() + startInd, labels.begin() + startInd + batchSize);
x = tx; //cerr << "tx=" << Debug(tx.shape()) << endl;
y = ty; //cerr << "ty=" << Debug(ty.shape()) << endl;
graph.forward(500); x = tx;
y = ty;
std::cerr << "z: " << Debug(z.val().shape()) << endl; cerr << "x=" << Debug(x.val().shape()) << endl;
std::cerr << "lr: " << Debug(lr.val().shape()) << endl; cerr << "y=" << Debug(y.val().shape()) << endl;
std::cerr << "Log-likelihood: " << Debug(graph.val().shape()) << endl ;
//std::cerr << "scores=" << scores.val().Debug() << endl;
std::cerr << "lr=" << lr.val().Debug() << endl;
graph.backward(); graph.forward(batchSize);
//std::cerr << graph["pred"].val()[0] << std::endl; cerr << "w=" << Debug(w.val().shape()) << endl;
cerr << "b=" << Debug(b.val().shape()) << endl;
std::cerr << "z: " << Debug(z.val().shape()) << endl;
std::cerr << "lr: " << Debug(lr.val().shape()) << endl;
std::cerr << "Log-likelihood: " << Debug(graph.val().shape()) << endl ;
//std::cerr << "scores=" << scores.val().Debug() << endl;
std::cerr << "lr=" << lr.val().Debug() << endl;
graph.backward();
//std::cerr << graph["pred"].val()[0] << std::endl;
startInd += batchSize;
startIndData += batchSize * IMAGE_SIZE;
}
// XOR // XOR