mirror of
https://github.com/marian-nmt/marian.git
synced 2024-09-11 06:15:56 +03:00
Merge branch 'master' of github.com:emjotde/Marian
Conflicts: src/validate_mnist.cu
This commit is contained in:
commit
4e63816d60
@ -56,11 +56,11 @@
|
||||
</tool>
|
||||
</toolChain>
|
||||
</folderInfo>
|
||||
<fileInfo id="com.nvidia.cuda.ide.seven_five.configuration.debug.1479727693.1770712140" name="tensor.cu" rcbsApplicability="disable" resourcePath="tensor.cu" toolsToInvoke="nvcc.compiler.base.1979453423.1452902875">
|
||||
<tool id="nvcc.compiler.base.1979453423.1452902875" name="NVCC Compiler" superClass="nvcc.compiler.base.1979453423"/>
|
||||
<fileInfo id="com.nvidia.cuda.ide.seven_five.configuration.debug.1479727693.924444438" name="train_mnist.cu" rcbsApplicability="disable" resourcePath="src/train_mnist.cu" toolsToInvoke="nvcc.compiler.base.1979453423.2078504098">
|
||||
<tool id="nvcc.compiler.base.1979453423.2078504098" name="NVCC Compiler" superClass="nvcc.compiler.base.1979453423"/>
|
||||
</fileInfo>
|
||||
<sourceEntries>
|
||||
<entry flags="VALUE_WORKSPACE_PATH|RESOLVED" kind="sourcePath" name=""/>
|
||||
<entry excluding="src/train_mnist.cu|src/validate_mnist.cu|src/npz_converter.cpp" flags="VALUE_WORKSPACE_PATH|RESOLVED" kind="sourcePath" name=""/>
|
||||
</sourceEntries>
|
||||
</configuration>
|
||||
</storageModule>
|
||||
|
72
scripts/train_test_model_multi.py
Executable file
72
scripts/train_test_model_multi.py
Executable file
@ -0,0 +1,72 @@
|
||||
#!/usr/bin/env python
|
||||
|
||||
import sys
|
||||
import os
|
||||
import numpy as np
|
||||
from keras.datasets import mnist
|
||||
from keras.utils import np_utils
|
||||
from keras.models import Sequential
|
||||
from keras.layers import Dense
|
||||
from keras.layers import Dropout
|
||||
|
||||
def softmax(x):
|
||||
return np.exp(x) / np.sum(np.exp(x), axis=1)[:, None]
|
||||
|
||||
|
||||
def baseline_model(pixels_count, classes_count):
|
||||
model = Sequential()
|
||||
model.add(Dense(100, input_dim=pixels_count, init='normal', activation='tanh'))
|
||||
model.add(Dense(classes_count, input_dim=100, init='normal', activation='softmax'))
|
||||
model.compile(loss='categorical_crossentropy', optimizer='adam', metrics=['accuracy'])
|
||||
return model
|
||||
|
||||
|
||||
if __name__ == "__main__":
|
||||
### Load trainset from mnist
|
||||
|
||||
(X_train, y_train), (X_test, y_test) = mnist.load_data()
|
||||
|
||||
### Flatten pictures into vectors
|
||||
|
||||
pixels_count = X_train.shape[1] * X_train.shape[2]
|
||||
X_train = X_train.reshape(X_train.shape[0], pixels_count).astype('float32')
|
||||
print "X shape: ", X_train.shape
|
||||
|
||||
X_test = X_test.reshape(X_test.shape[0], pixels_count).astype('float32')
|
||||
|
||||
### Normalize data to (0, 1)
|
||||
|
||||
X_train = X_train / 255
|
||||
X_test = X_test / 255
|
||||
|
||||
### Change classes to one hot encoding matrixes
|
||||
|
||||
y_train = np_utils.to_categorical(y_train)
|
||||
classes_count = y_train.shape[1]
|
||||
print "Y shape: ", y_train.shape
|
||||
|
||||
y_test = np_utils.to_categorical(y_test)
|
||||
|
||||
# Train weight matrix
|
||||
|
||||
# Build the model
|
||||
model = baseline_model(pixels_count, classes_count)
|
||||
# Fit the model
|
||||
model.fit(X_train, y_train, validation_data=(X_test, y_test), nb_epoch=10, batch_size=200, verbose=2)
|
||||
# Final evaluation of the model
|
||||
scores = model.evaluate(X_test, y_test, verbose=0)
|
||||
print("Baseline Error: %.2f%%" % (100-scores[1]*100))
|
||||
|
||||
### Weight and bias matrixes - we extract them from the model
|
||||
|
||||
# weights_ones = np.ones((pixels_count, classes_count))
|
||||
# print weights_ones.shape
|
||||
|
||||
weights1, bias1, weights2, bias2 = model.get_weights()
|
||||
### Save model to npz files
|
||||
if not os.path.exists("test_model_multi"):
|
||||
os.makedirs("test_model_multi")
|
||||
# np.savez("test_model_multi/model", *model)
|
||||
np.savez("test_model_multi/model", weights1 = weights1, bias1 = bias1, weights2 = weights2, bias2 = bias2)
|
||||
|
||||
print "Model saved! Check test_model_multi directory"
|
@ -84,8 +84,8 @@ if __name__ == "__main__":
|
||||
# print np.count_nonzero(lr)i
|
||||
|
||||
### Save model to npz files
|
||||
if not os.path.exists("test_model"):
|
||||
os.makedirs("test_model")
|
||||
np.savez("test_model/model", weights = weights, bias = bias)
|
||||
if not os.path.exists("test_model_single"):
|
||||
os.makedirs("test_model_single")
|
||||
np.savez("test_model_single/model", weights = weights, bias = bias)
|
||||
|
||||
print "Model saved! Check test_model directory"
|
||||
print "Model saved! Check test_model_single directory"
|
@ -1,39 +1,43 @@
|
||||
|
||||
include_directories(.)
|
||||
|
||||
add_library(libcommon OBJECT
|
||||
exception.cpp
|
||||
cuda_add_library(marian_lib
|
||||
cnpy/cnpy.cpp
|
||||
exception.cpp
|
||||
expressions.cu
|
||||
tensor.cu
|
||||
tensor_operators.cu
|
||||
)
|
||||
|
||||
target_link_libraries(marian_lib)
|
||||
|
||||
cuda_add_executable(
|
||||
marian
|
||||
test.cu
|
||||
expressions.cu
|
||||
tensor_operators.cu
|
||||
tensor.cu
|
||||
$<TARGET_OBJECTS:libcommon>
|
||||
)
|
||||
|
||||
target_link_libraries(marian marian_lib)
|
||||
|
||||
cuda_add_executable(
|
||||
train_mnist
|
||||
train_mnist.cu
|
||||
expressions.cu
|
||||
tensor_operators.cu
|
||||
tensor.cu
|
||||
$<TARGET_OBJECTS:libcommon>
|
||||
)
|
||||
|
||||
target_link_libraries(train_mnist marian_lib)
|
||||
|
||||
cuda_add_executable(
|
||||
validate_mnist
|
||||
validate_mnist.cu
|
||||
expressions.cu
|
||||
tensor_operators.cu
|
||||
tensor.cu
|
||||
$<TARGET_OBJECTS:libcommon>
|
||||
)
|
||||
cuda_add_executable(
|
||||
validate_mnist_batch
|
||||
validate_mnist_batch.cu
|
||||
)
|
||||
|
||||
foreach(exec marian train_mnist validate_mnist)
|
||||
target_link_libraries(validate_mnist marian_lib)
|
||||
target_link_libraries(validate_mnist_batch marian_lib)
|
||||
|
||||
foreach(exec marian train_mnist validate_mnist validate_mnist_batch )
|
||||
target_link_libraries(${exec} ${EXT_LIBS} cuda cudnn)
|
||||
cuda_add_cublas_to_target(${exec})
|
||||
set_target_properties(${exec} PROPERTIES RUNTIME_OUTPUT_DIRECTORY "${CMAKE_BINARY_DIR}")
|
||||
|
@ -5,4 +5,5 @@
|
||||
#include "graph_operators.h"
|
||||
#include "expressions.h"
|
||||
#include "expression_operators.h"
|
||||
#include "param_initializers.h"
|
||||
|
||||
|
@ -1,4 +1,4 @@
|
||||
#include "common/npz_converter.h"
|
||||
#include "npz_converter.h"
|
||||
|
||||
|
||||
|
||||
|
34
src/param_initializers.h
Normal file
34
src/param_initializers.h
Normal file
@ -0,0 +1,34 @@
|
||||
#pragma once
|
||||
|
||||
#include <random>
|
||||
#include <algorithm>
|
||||
#include <iterator>
|
||||
#include <functional>
|
||||
|
||||
#include "tensor.h"
|
||||
|
||||
namespace marian {
|
||||
|
||||
void zeros(Tensor t) {
|
||||
std::vector<float> vals(t.size(), 0.0f);
|
||||
thrust::copy(vals.begin(), vals.end(), t.begin());
|
||||
}
|
||||
|
||||
void ones(Tensor t) {
|
||||
std::vector<float> vals(t.size(), 1.0f);
|
||||
thrust::copy(vals.begin(), vals.end(), t.begin());
|
||||
}
|
||||
|
||||
void randreal(Tensor t) {
|
||||
std::random_device device;
|
||||
std::default_random_engine engine(device());
|
||||
std::uniform_real_distribution<> dist(0, 1);
|
||||
auto gen = std::bind(dist, engine);
|
||||
|
||||
std::vector<float> vals(t.size());
|
||||
std::generate(begin(vals), end(vals), gen);
|
||||
|
||||
thrust::copy(vals.begin(), vals.end(), t.begin());
|
||||
}
|
||||
|
||||
} // namespace marian
|
@ -60,8 +60,8 @@ class SGD {
|
||||
std::vector<float> y(yData_.begin() + startId * numClasses_,
|
||||
yData_.begin() + endId * numClasses_);
|
||||
|
||||
xt.Load(x);
|
||||
yt.Load(y);
|
||||
xt.set(x);
|
||||
yt.set(y);
|
||||
}
|
||||
|
||||
void UpdateModel() {
|
||||
|
@ -5,94 +5,26 @@ using namespace std;
|
||||
|
||||
namespace marian {
|
||||
|
||||
inline std::vector<std::string> Tokenize(const std::string& str,
|
||||
const std::string& delimiters = " \t")
|
||||
{
|
||||
std::vector<std::string> tokens;
|
||||
// Skip delimiters at beginning.
|
||||
std::string::size_type lastPos = str.find_first_not_of(delimiters, 0);
|
||||
// Find first "non-delimiter".
|
||||
std::string::size_type pos = str.find_first_of(delimiters, lastPos);
|
||||
|
||||
while (std::string::npos != pos || std::string::npos != lastPos) {
|
||||
// Found a token, add it to the vector.
|
||||
tokens.push_back(str.substr(lastPos, pos - lastPos));
|
||||
// Skip delimiters. Note the "not_of"
|
||||
lastPos = str.find_first_not_of(delimiters, pos);
|
||||
// Find next "non-delimiter"
|
||||
pos = str.find_first_of(delimiters, lastPos);
|
||||
}
|
||||
|
||||
return tokens;
|
||||
}
|
||||
|
||||
//! convert string to variable of type T. Used to reading floats, int etc from files
|
||||
template<typename T>
|
||||
T Scan(const std::string &input)
|
||||
{
|
||||
std::stringstream stream(input);
|
||||
T ret;
|
||||
stream >> ret;
|
||||
return ret;
|
||||
}
|
||||
|
||||
//! convert vectors of string to vectors of type T variables
|
||||
template<typename T>
|
||||
inline std::vector<T> Scan(const std::vector< std::string > &input)
|
||||
{
|
||||
std::vector<T> output(input.size());
|
||||
for (size_t i = 0 ; i < input.size() ; i++) {
|
||||
output[i] = Scan<T>( input[i] );
|
||||
}
|
||||
return output;
|
||||
}
|
||||
|
||||
//! tokenise input string to vector of type T
|
||||
template<typename T>
|
||||
inline std::vector<T> Tokenize( const std::string &input
|
||||
, const std::string& delimiters = " \t")
|
||||
{
|
||||
std::vector<std::string> stringVector = Tokenize(input, delimiters);
|
||||
return Scan<T>( stringVector );
|
||||
}
|
||||
|
||||
|
||||
void Tensor::Load(const std::string &path)
|
||||
{
|
||||
size_t totSize = GetTotalSize(pimpl_->shape());
|
||||
cerr << "totSize=" << totSize << endl;
|
||||
std::vector<float> hostData(totSize);
|
||||
|
||||
fstream strm;
|
||||
strm.open(path.c_str());
|
||||
|
||||
string line;
|
||||
size_t ind = 0;
|
||||
while ( getline (strm, line) )
|
||||
{
|
||||
cerr << line << '\n';
|
||||
vector<Float> toks = Tokenize<Float>(line);
|
||||
for (size_t i = 0; i < toks.size(); ++i) {
|
||||
hostData[ind] = toks[i];
|
||||
}
|
||||
|
||||
++ind;
|
||||
}
|
||||
strm.close();
|
||||
|
||||
Load(hostData.begin(), hostData.begin());
|
||||
}
|
||||
|
||||
void Tensor::Load(const std::vector<float>& data)
|
||||
void Tensor::set(const std::vector<float>& data)
|
||||
{
|
||||
pimpl_->set(data.begin(), data.end());
|
||||
}
|
||||
|
||||
|
||||
void Tensor::Load(const std::vector<float>::const_iterator &begin, const std::vector<float>::const_iterator &end)
|
||||
void Tensor::set(const std::vector<float>::const_iterator &begin, const std::vector<float>::const_iterator &end)
|
||||
{
|
||||
pimpl_->set(begin, end);
|
||||
}
|
||||
|
||||
Tensor& operator<<(Tensor& t, const std::vector<float> &vec) {
|
||||
t.set(vec);
|
||||
return t;
|
||||
}
|
||||
|
||||
std::vector<float>& operator<<(std::vector<float> &vec, const Tensor& t) {
|
||||
t.get(vec);
|
||||
return vec;
|
||||
}
|
||||
|
||||
|
||||
}
|
||||
|
||||
|
85
src/tensor.h
85
src/tensor.h
@ -1,6 +1,5 @@
|
||||
#pragma once
|
||||
|
||||
#include <cudnn.h>
|
||||
#include <cublas_v2.h>
|
||||
#include <thrust/device_vector.h>
|
||||
#include <thrust/functional.h>
|
||||
@ -13,30 +12,6 @@
|
||||
|
||||
namespace marian {
|
||||
|
||||
struct Handles {
|
||||
cudnnHandle_t cudnnHandle;
|
||||
cublasHandle_t cublasHandle;
|
||||
|
||||
cudnnOpTensorDescriptor_t add;
|
||||
|
||||
Handles() {
|
||||
cudnnCreate(&cudnnHandle);
|
||||
cublasCreate(&cublasHandle);
|
||||
cudnnCreateOpTensorDescriptor(&add);
|
||||
cudnnSetOpTensorDescriptor(add, CUDNN_OP_TENSOR_ADD, CUDNN_DATA_FLOAT, CUDNN_NOT_PROPAGATE_NAN);
|
||||
}
|
||||
|
||||
~Handles() {
|
||||
cudnnDestroy(cudnnHandle);
|
||||
cublasDestroy(cublasHandle);
|
||||
cudnnDestroyOpTensorDescriptor(add);
|
||||
}
|
||||
};
|
||||
|
||||
const Handles handles;
|
||||
|
||||
// typedef std::vector<int> Shape;
|
||||
|
||||
inline std::string Debug(const Shape &shape)
|
||||
{
|
||||
std::stringstream strm;
|
||||
@ -60,18 +35,9 @@ class TensorImpl {
|
||||
private:
|
||||
Shape shape_;
|
||||
thrust::device_vector<Float> data_;
|
||||
cudnnTensorDescriptor_t desc_;
|
||||
size_t tno_;
|
||||
static size_t tensorCounter;
|
||||
|
||||
cudnnDataType_t dataType() {
|
||||
switch(sizeof(Float)) {
|
||||
case 2: return CUDNN_DATA_HALF;
|
||||
case 8: return CUDNN_DATA_DOUBLE;
|
||||
default: return CUDNN_DATA_FLOAT;
|
||||
}
|
||||
}
|
||||
|
||||
public:
|
||||
typedef Float value_type;
|
||||
|
||||
@ -90,30 +56,11 @@ class TensorImpl {
|
||||
|
||||
int size = GetTotalSize(shape_);
|
||||
data_.resize(size, value);
|
||||
cudnnCreateTensorDescriptor(&desc_);
|
||||
switch (shape_.size()) {
|
||||
case 1:
|
||||
cudnnSetTensor4dDescriptor(desc_, CUDNN_TENSOR_NCHW, dataType(),
|
||||
shape_[0], 1, 1, 1); break;
|
||||
case 2:
|
||||
cudnnSetTensor4dDescriptor(desc_, CUDNN_TENSOR_NCHW, dataType(),
|
||||
shape_[0], shape_[1], 1, 1); break;
|
||||
case 3:
|
||||
cudnnSetTensor4dDescriptor(desc_, CUDNN_TENSOR_NCHW, dataType(),
|
||||
shape_[0], shape_[1], shape_[2], 1); break;
|
||||
case 4:
|
||||
cudnnSetTensor4dDescriptor(desc_, CUDNN_TENSOR_NCHW, dataType(),
|
||||
shape_[0], shape_[1], shape_[2], shape_[3]); break;
|
||||
}
|
||||
}
|
||||
|
||||
TensorImpl(const TensorImpl&) = delete;
|
||||
TensorImpl(TensorImpl&&) = delete;
|
||||
|
||||
~TensorImpl() {
|
||||
cudnnDestroyTensorDescriptor(desc_);
|
||||
}
|
||||
|
||||
value_type operator[](size_t i) const {
|
||||
return data_[i];
|
||||
}
|
||||
@ -146,10 +93,6 @@ class TensorImpl {
|
||||
return thrust::raw_pointer_cast(data_.data());
|
||||
}
|
||||
|
||||
cudnnTensorDescriptor_t desc() const {
|
||||
return desc_;
|
||||
}
|
||||
|
||||
size_t id() const {
|
||||
return tno_;
|
||||
}
|
||||
@ -159,12 +102,13 @@ class TensorImpl {
|
||||
}
|
||||
|
||||
void set(const std::vector<float>::const_iterator &begin, const std::vector<float>::const_iterator &end) {
|
||||
size_t totSize = GetTotalSize(shape());
|
||||
//std::cerr << "tensor size=" << totSize << " vector size=" << values.size() << std::endl;
|
||||
//assert(totSize == values.size());
|
||||
thrust::copy(begin, end, data_.begin());
|
||||
}
|
||||
|
||||
void get(std::vector<float>::iterator out) {
|
||||
thrust::copy(data_.begin(), data_.end(), out);
|
||||
}
|
||||
|
||||
std::string Debug() const
|
||||
{
|
||||
std::stringstream strm;
|
||||
@ -246,10 +190,6 @@ class Tensor {
|
||||
return pimpl_->shape();
|
||||
}
|
||||
|
||||
cudnnTensorDescriptor_t desc() const {
|
||||
return pimpl_->desc();
|
||||
}
|
||||
|
||||
void set(value_type value) {
|
||||
pimpl_->set(value);
|
||||
}
|
||||
@ -274,10 +214,21 @@ class Tensor {
|
||||
std::cerr << std::endl;
|
||||
}
|
||||
|
||||
void Load(const std::string &path);
|
||||
void Load(const std::vector<float>& data);
|
||||
void Load(const std::vector<float>::const_iterator &begin, const std::vector<float>::const_iterator &end);
|
||||
//void Load(const std::string &path);
|
||||
void set(const std::vector<float>& data);
|
||||
void set(const std::vector<float>::const_iterator &begin, const std::vector<float>::const_iterator &end);
|
||||
|
||||
void get(std::vector<float>::iterator out) const {
|
||||
pimpl_->get(out);
|
||||
}
|
||||
|
||||
void get(std::vector<float> &vout) const {
|
||||
pimpl_->get(vout.begin());
|
||||
}
|
||||
};
|
||||
|
||||
Tensor& operator<<(Tensor& t, const std::vector<float> &vec);
|
||||
|
||||
std::vector<float>& operator<<(std::vector<float> &vec, const Tensor& t);
|
||||
|
||||
}
|
||||
|
@ -130,7 +130,11 @@ Tensor Prod(cublasHandle_t handle, Tensor C, const Tensor A, const Tensor B,
|
||||
Tensor Prod(Tensor C, const Tensor A, const Tensor B,
|
||||
bool transA, bool transB, Float beta) {
|
||||
|
||||
return Prod(handles.cublasHandle, C, A, B, transA, transB, beta);
|
||||
cublasHandle_t cublasHandle;
|
||||
cublasCreate(&cublasHandle);
|
||||
Tensor temp = Prod(cublasHandle, C, A, B, transA, transB, beta);
|
||||
cublasDestroy(cublasHandle);
|
||||
return temp;
|
||||
}
|
||||
|
||||
}
|
207
src/test.cu
207
src/test.cu
@ -2,9 +2,9 @@
|
||||
#include "marian.h"
|
||||
#include "mnist.h"
|
||||
|
||||
using namespace std;
|
||||
|
||||
int main(int argc, char** argv) {
|
||||
cudaSetDevice(0);
|
||||
|
||||
/*int numImg = 0;*/
|
||||
/*auto images = datasets::mnist::ReadImages("../examples/mnist/t10k-images-idx3-ubyte", numImg);*/
|
||||
/*auto labels = datasets::mnist::ReadLabels("../examples/mnist/t10k-labels-idx1-ubyte", numImg);*/
|
||||
@ -12,117 +12,104 @@ int main(int argc, char** argv) {
|
||||
using namespace marian;
|
||||
using namespace keywords;
|
||||
|
||||
const size_t BATCH_SIZE = 500;
|
||||
const size_t IMAGE_SIZE = 784;
|
||||
const size_t LABEL_SIZE = 10;
|
||||
|
||||
Expr x = input(shape={whatevs, IMAGE_SIZE}, name="X");
|
||||
Expr y = input(shape={whatevs, LABEL_SIZE}, name="Y");
|
||||
Expr x = input(shape={1, 2});
|
||||
Expr y = input(shape={1, 2});
|
||||
|
||||
Expr w = param(shape={IMAGE_SIZE, LABEL_SIZE}, name="W0");
|
||||
Expr b = param(shape={1, LABEL_SIZE}, name="b0");
|
||||
|
||||
Expr z = dot(x, w) + b;
|
||||
Expr lr = softmax(z, axis=1, name="pred");
|
||||
Expr graph = -mean(sum(y * log(lr), axis=1), axis=0, name="cost");
|
||||
//cerr << "x=" << Debug(lr.val().shape()) << endl;
|
||||
Expr w = param(shape={2, 2}, name="W0");
|
||||
//Expr b = param(shape={1, 2}, name="b0");
|
||||
|
||||
int numofdata;
|
||||
//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> 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 << "numofdata=" << numofdata << endl;
|
||||
|
||||
size_t startInd = 0;
|
||||
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;
|
||||
|
||||
Tensor tx({numofdata, IMAGE_SIZE}, 1);
|
||||
Tensor ty({numofdata, LABEL_SIZE}, 1);
|
||||
|
||||
tx.Load(images.begin() + startIndData, images.begin() + startIndData + batchSize * IMAGE_SIZE);
|
||||
ty.Load(labels.begin() + startInd, labels.begin() + startInd + batchSize);
|
||||
|
||||
//cerr << "tx=" << Debug(tx.shape()) << endl;
|
||||
//cerr << "ty=" << Debug(ty.shape()) << endl;
|
||||
|
||||
x = tx;
|
||||
y = ty;
|
||||
|
||||
cerr << "x=" << Debug(x.val().shape()) << endl;
|
||||
cerr << "y=" << Debug(y.val().shape()) << endl;
|
||||
|
||||
|
||||
graph.forward(batchSize);
|
||||
|
||||
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;
|
||||
}
|
||||
std::cerr << "Building model...";
|
||||
auto predict = softmax(dot(x, w),
|
||||
axis=1, name="pred");
|
||||
auto graph = -mean(sum(y * log(predict), axis=1),
|
||||
axis=0, name="cost");
|
||||
|
||||
|
||||
// XOR
|
||||
/*
|
||||
Expr x = input(shape={whatevs, 2}, name="X");
|
||||
Expr y = input(shape={whatevs, 2}, name="Y");
|
||||
|
||||
Expr w = param(shape={2, 1}, name="W0");
|
||||
Expr b = param(shape={1, 1}, name="b0");
|
||||
|
||||
Expr n5 = dot(x, w);
|
||||
Expr n6 = n5 + b;
|
||||
Expr lr = softmax(n6, axis=1, name="pred");
|
||||
cerr << "lr=" << lr.Debug() << endl;
|
||||
|
||||
Expr graph = -mean(sum(y * log(lr), axis=1), axis=0, name="cost");
|
||||
|
||||
Tensor tx({4, 2}, 1);
|
||||
Tensor ty({4, 1}, 1);
|
||||
cerr << "tx=" << tx.Debug() << endl;
|
||||
cerr << "ty=" << ty.Debug() << endl;
|
||||
|
||||
tx.Load("../examples/xor/train.txt");
|
||||
ty.Load("../examples/xor/label.txt");
|
||||
*/
|
||||
|
||||
#if 0
|
||||
hook0(graph);
|
||||
graph.autodiff();
|
||||
std::cerr << graph["cost"].val()[0] << std::endl;
|
||||
//hook1(graph);
|
||||
for(auto p : graph.params()) {
|
||||
auto update = _1 = _1 - alpha * _2;
|
||||
Element(update, p.val(), p.grad());
|
||||
}
|
||||
hook2(graph);
|
||||
Tensor x1t({1, 2});
|
||||
std::vector<float> xv = { 0.6, 0.1 };
|
||||
thrust::copy(xv.begin(), xv.end(), x1t.begin());
|
||||
|
||||
Tensor x2t({1, 2});
|
||||
std::vector<float> yv = { 0, 1 };
|
||||
thrust::copy(yv.begin(), yv.end(), x2t.begin());
|
||||
|
||||
x = x1t;
|
||||
y = x2t;
|
||||
|
||||
graph.forward(1);
|
||||
graph.backward();
|
||||
|
||||
std::cerr << graph.val().Debug() << std::endl;
|
||||
std::cerr << w.grad().Debug() << std::endl;
|
||||
//std::cerr << b.grad().Debug() << std::endl;
|
||||
|
||||
// using namespace marian;
|
||||
// using namespace keywords;
|
||||
//
|
||||
// const size_t BATCH_SIZE = 500;
|
||||
// const size_t IMAGE_SIZE = 784;
|
||||
// const size_t LABEL_SIZE = 10;
|
||||
//
|
||||
// Expr x = input(shape={whatevs, IMAGE_SIZE}, name="X");
|
||||
// Expr y = input(shape={whatevs, LABEL_SIZE}, name="Y");
|
||||
//
|
||||
// Expr w = param(shape={IMAGE_SIZE, LABEL_SIZE}, name="W0");
|
||||
// Expr b = param(shape={1, LABEL_SIZE}, name="b0");
|
||||
//
|
||||
// Expr z = dot(x, w) + b;
|
||||
// Expr lr = softmax(z, axis=1, name="pred");
|
||||
// Expr graph = -mean(sum(y * log(lr), axis=1), axis=0, name="cost");
|
||||
// //cerr << "x=" << Debug(lr.val().shape()) << endl;
|
||||
//
|
||||
// int numofdata;
|
||||
// //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> 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 << "numofdata=" << numofdata << endl;
|
||||
//
|
||||
// size_t startInd = 0;
|
||||
// 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;
|
||||
//
|
||||
// Tensor tx({numofdata, IMAGE_SIZE}, 1);
|
||||
// Tensor ty({numofdata, LABEL_SIZE}, 1);
|
||||
//
|
||||
// tx.Load(images.begin() + startIndData, images.begin() + startIndData + batchSize * IMAGE_SIZE);
|
||||
// ty.Load(labels.begin() + startInd, labels.begin() + startInd + batchSize);
|
||||
//
|
||||
// //cerr << "tx=" << Debug(tx.shape()) << endl;
|
||||
// //cerr << "ty=" << Debug(ty.shape()) << endl;
|
||||
//
|
||||
// x = tx;
|
||||
// y = ty;
|
||||
//
|
||||
// cerr << "x=" << Debug(x.val().shape()) << endl;
|
||||
// cerr << "y=" << Debug(y.val().shape()) << endl;
|
||||
//
|
||||
//
|
||||
// graph.forward(batchSize);
|
||||
//
|
||||
// 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;
|
||||
// }
|
||||
|
||||
auto opt = adadelta(cost_function=cost,
|
||||
eta=0.9, gamma=0.1,
|
||||
set_batch=set,
|
||||
before_update=before,
|
||||
after_update=after,
|
||||
set_valid=valid,
|
||||
validation_freq=100,
|
||||
verbose=1, epochs=3, early_stopping=10);
|
||||
opt.run();
|
||||
#endif
|
||||
return 0;
|
||||
}
|
||||
|
@ -7,104 +7,69 @@ using namespace marian;
|
||||
using namespace keywords;
|
||||
|
||||
int main(int argc, char** argv) {
|
||||
|
||||
cudaSetDevice(0);
|
||||
|
||||
const size_t IMAGE_SIZE = 784;
|
||||
const size_t LABEL_SIZE = 10;
|
||||
const size_t BATCH_SIZE = 24;
|
||||
int numofdata;
|
||||
|
||||
int BATCH_SIZE = 10000;
|
||||
|
||||
std::cerr << "Loading test set...";
|
||||
std::vector<float> testImages = datasets::mnist::ReadImages("../examples/mnist/t10k-images-idx3-ubyte", numofdata, IMAGE_SIZE);
|
||||
std::vector<float> testLabels = datasets::mnist::ReadLabels("../examples/mnist/t10k-labels-idx1-ubyte", numofdata, LABEL_SIZE);
|
||||
std::cerr << "\tDone." << std::endl;
|
||||
std::vector<float> testImages = datasets::mnist::ReadImages("../examples/mnist/t10k-images-idx3-ubyte", BATCH_SIZE, IMAGE_SIZE);
|
||||
std::vector<float> testLabels = datasets::mnist::ReadLabels("../examples/mnist/t10k-labels-idx1-ubyte", BATCH_SIZE, LABEL_SIZE);
|
||||
std::cerr << "Done." << std::endl;
|
||||
|
||||
|
||||
std::cerr << "Loading model params...";
|
||||
NpzConverter converter("../scripts/test_model/model.npz");
|
||||
|
||||
std::vector<float> wData;
|
||||
Shape wShape;
|
||||
std::vector<float> wData, bData;
|
||||
Shape wShape, bShape;
|
||||
converter.Load("weights", wData, wShape);
|
||||
|
||||
std::vector<float> bData;
|
||||
Shape bShape;
|
||||
converter.Load("bias", bData, bShape);
|
||||
|
||||
auto initW = [&wData](Tensor t) {
|
||||
thrust::copy(wData.begin(), wData.end(), t.begin());
|
||||
};
|
||||
|
||||
auto initB = [&bData](Tensor t) {
|
||||
thrust::copy(bData.begin(), bData.end(), t.begin());
|
||||
};
|
||||
|
||||
std::cerr << "\tDone." << std::endl;
|
||||
|
||||
|
||||
Expr x = input(shape={whatevs, IMAGE_SIZE}, name="X");
|
||||
|
||||
Expr w = param(shape={IMAGE_SIZE, LABEL_SIZE}, name="W0", init=initW);
|
||||
Expr b = param(shape={1, LABEL_SIZE}, name="b0", init=initB);
|
||||
std::cerr << "Done." << std::endl;
|
||||
|
||||
std::cerr << "Building model...";
|
||||
auto scores = dot(x, w) + b;
|
||||
auto predict = softmax(scores, axis=1, name="pred");
|
||||
std::cerr << "\tDone." << std::endl;
|
||||
auto x = input(shape={whatevs, IMAGE_SIZE}, name="X");
|
||||
auto y = input(shape={whatevs, LABEL_SIZE}, name="Y");
|
||||
|
||||
auto w = param(shape={IMAGE_SIZE, LABEL_SIZE}, name="W0",
|
||||
init=[wData](Tensor t) { t.set(wData); });
|
||||
auto b = param(shape={1, LABEL_SIZE}, name="b0",
|
||||
init=[bData](Tensor t) {t.set(bData); });
|
||||
|
||||
auto predict = softmax(dot(x, w) + b,
|
||||
axis=1, name="pred");
|
||||
auto graph = -mean(sum(y * log(predict), axis=1),
|
||||
axis=0, name="cost");
|
||||
std::cerr << "Done." << std::endl;
|
||||
|
||||
Tensor xt({BATCH_SIZE, IMAGE_SIZE});
|
||||
/* xt.Load(testImages); */
|
||||
/* x = xt; */
|
||||
Tensor yt({BATCH_SIZE, LABEL_SIZE});
|
||||
|
||||
x = xt << testImages;
|
||||
y = yt << testLabels;
|
||||
|
||||
graph.forward(BATCH_SIZE);
|
||||
graph.backward();
|
||||
|
||||
auto results = predict.val();
|
||||
std::vector<float> resultsv(results.size());
|
||||
resultsv << results;
|
||||
|
||||
std::cerr << b.grad().Debug() << std::endl;
|
||||
|
||||
size_t acc = 0;
|
||||
size_t startId = 0;
|
||||
size_t endId = startId + BATCH_SIZE;
|
||||
|
||||
while (endId < numofdata) {
|
||||
std::vector<float> tmp(testImages.begin() + (startId * IMAGE_SIZE),
|
||||
testImages.begin() + (endId * IMAGE_SIZE));
|
||||
xt.Load(tmp);
|
||||
x = xt;
|
||||
|
||||
predict.forward(BATCH_SIZE);
|
||||
|
||||
thrust::host_vector<float> results(predict.val().begin(), predict.val().begin() + LABEL_SIZE * BATCH_SIZE);
|
||||
|
||||
for (size_t i = 0; i < BATCH_SIZE * LABEL_SIZE; i += LABEL_SIZE) {
|
||||
size_t correct = 0;
|
||||
size_t predicted = 0;
|
||||
for (size_t j = 0; j < LABEL_SIZE; ++j) {
|
||||
if (testLabels[startId * LABEL_SIZE + i + j]) correct = j;
|
||||
if (results[i + j] > results[i + predicted]) predicted = j;
|
||||
}
|
||||
acc += (correct == predicted);
|
||||
for (size_t i = 0; i < testLabels.size(); i += LABEL_SIZE) {
|
||||
size_t correct = 0;
|
||||
size_t predicted = 0;
|
||||
for (size_t j = 0; j < LABEL_SIZE; ++j) {
|
||||
if (testLabels[i+j]) correct = j;
|
||||
if (resultsv[i + j] > resultsv[i + predicted]) predicted = j;
|
||||
}
|
||||
|
||||
startId += BATCH_SIZE;
|
||||
endId += BATCH_SIZE;
|
||||
acc += (correct == predicted);
|
||||
}
|
||||
if (endId != numofdata) {
|
||||
endId = numofdata;
|
||||
if (endId - startId >= 0) {
|
||||
std::vector<float> tmp(testImages.begin() + (startId * IMAGE_SIZE),
|
||||
testImages.begin() + (endId * IMAGE_SIZE));
|
||||
xt.Load(tmp);
|
||||
x = xt;
|
||||
|
||||
predict.forward(endId - startId);
|
||||
|
||||
thrust::host_vector<float> results(predict.val().begin(), predict.val().begin() + LABEL_SIZE * (endId - startId));
|
||||
|
||||
for (size_t i = 0; i < (endId - startId) * LABEL_SIZE; i += LABEL_SIZE) {
|
||||
size_t correct = 0;
|
||||
size_t predicted = 0;
|
||||
for (size_t j = 0; j < LABEL_SIZE; ++j) {
|
||||
if (testLabels[startId * LABEL_SIZE + i + j]) correct = j;
|
||||
if (results[i + j] > results[i + predicted]) predicted = j;
|
||||
}
|
||||
acc += (correct == predicted);
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
std::cerr << "ACC: " << float(acc)/numofdata << std::endl;
|
||||
std::cerr << "Accuracy: " << float(acc)/BATCH_SIZE << std::endl;
|
||||
|
||||
return 0;
|
||||
}
|
||||
|
113
src/validate_mnist_batch.cu
Normal file
113
src/validate_mnist_batch.cu
Normal file
@ -0,0 +1,113 @@
|
||||
|
||||
#include "marian.h"
|
||||
#include "mnist.h"
|
||||
#include "npz_converter.h"
|
||||
|
||||
using namespace marian;
|
||||
using namespace keywords;
|
||||
|
||||
int main(int argc, char** argv) {
|
||||
|
||||
cudaSetDevice(0);
|
||||
|
||||
const size_t IMAGE_SIZE = 784;
|
||||
const size_t LABEL_SIZE = 10;
|
||||
const size_t BATCH_SIZE = 24;
|
||||
int numofdata;
|
||||
|
||||
std::cerr << "Loading test set...";
|
||||
std::vector<float> testImages = datasets::mnist::ReadImages("../examples/mnist/t10k-images-idx3-ubyte", numofdata, IMAGE_SIZE);
|
||||
std::vector<float> testLabels = datasets::mnist::ReadLabels("../examples/mnist/t10k-labels-idx1-ubyte", numofdata, LABEL_SIZE);
|
||||
std::cerr << "\tDone." << std::endl;
|
||||
|
||||
std::cerr << "Loading model params...";
|
||||
NpzConverter converter("../scripts/test_model/model.npz");
|
||||
|
||||
std::vector<float> wData;
|
||||
Shape wShape;
|
||||
converter.Load("weights", wData, wShape);
|
||||
|
||||
std::vector<float> bData;
|
||||
Shape bShape;
|
||||
converter.Load("bias", bData, bShape);
|
||||
|
||||
auto initW = [wData](Tensor t) {
|
||||
t.set(wData);
|
||||
};
|
||||
|
||||
auto initB = [bData](Tensor t) {
|
||||
t.set(bData);
|
||||
};
|
||||
|
||||
std::cerr << "\tDone." << std::endl;
|
||||
|
||||
|
||||
auto x = input(shape={whatevs, IMAGE_SIZE}, name="X");
|
||||
auto y = input(shape={whatevs, LABEL_SIZE}, name="Y");
|
||||
|
||||
auto w = param(shape={IMAGE_SIZE, LABEL_SIZE}, name="W0", init=initW);
|
||||
auto b = param(shape={1, LABEL_SIZE}, name="b0", init=initB);
|
||||
|
||||
std::cerr << "Building model...";
|
||||
auto predict = softmax(dot(x, w) + b, axis=1, name="pred");
|
||||
|
||||
std::cerr << "Done." << std::endl;
|
||||
|
||||
Tensor xt({BATCH_SIZE, IMAGE_SIZE});
|
||||
|
||||
size_t acc = 0;
|
||||
size_t startId = 0;
|
||||
size_t endId = startId + BATCH_SIZE;
|
||||
|
||||
while (endId < numofdata) {
|
||||
std::vector<float> tmp(testImages.begin() + (startId * IMAGE_SIZE),
|
||||
testImages.begin() + (endId * IMAGE_SIZE));
|
||||
xt << tmp;
|
||||
x = xt;
|
||||
|
||||
predict.forward(BATCH_SIZE);
|
||||
|
||||
std::vector<float> results(LABEL_SIZE * BATCH_SIZE);
|
||||
results << predict.val();
|
||||
|
||||
for (size_t i = 0; i < BATCH_SIZE * LABEL_SIZE; i += LABEL_SIZE) {
|
||||
size_t correct = 0;
|
||||
size_t predicted = 0;
|
||||
for (size_t j = 0; j < LABEL_SIZE; ++j) {
|
||||
if (testLabels[startId * LABEL_SIZE + i + j]) correct = j;
|
||||
if (results[i + j] > results[i + predicted]) predicted = j;
|
||||
}
|
||||
acc += (correct == predicted);
|
||||
}
|
||||
|
||||
startId += BATCH_SIZE;
|
||||
endId += BATCH_SIZE;
|
||||
}
|
||||
if (endId != numofdata) {
|
||||
endId = numofdata;
|
||||
if (endId - startId > 0) {
|
||||
std::vector<float> tmp(testImages.begin() + (startId * IMAGE_SIZE),
|
||||
testImages.begin() + (endId * IMAGE_SIZE));
|
||||
xt << tmp;
|
||||
x = xt;
|
||||
|
||||
predict.forward(endId - startId);
|
||||
|
||||
std::vector<float> results(LABEL_SIZE * BATCH_SIZE);
|
||||
results << predict.val();
|
||||
|
||||
for (size_t i = 0; i < (endId - startId) * LABEL_SIZE; i += LABEL_SIZE) {
|
||||
size_t correct = 0;
|
||||
size_t predicted = 0;
|
||||
for (size_t j = 0; j < LABEL_SIZE; ++j) {
|
||||
if (testLabels[startId * LABEL_SIZE + i + j]) correct = j;
|
||||
if (results[i + j] > results[i + predicted]) predicted = j;
|
||||
}
|
||||
acc += (correct == predicted);
|
||||
}
|
||||
}
|
||||
}
|
||||
std::cerr << "ACC: " << float(acc)/numofdata << std::endl;
|
||||
|
||||
return 0;
|
||||
}
|
Loading…
Reference in New Issue
Block a user