788 lines
28 KiB
C++
788 lines
28 KiB
C++
/*
|
|
* <Short Description>
|
|
* Copyright (C) 2024 Brett Terpstra
|
|
*
|
|
* This program is free software: you can redistribute it and/or modify
|
|
* it under the terms of the GNU General Public License as published by
|
|
* the Free Software Foundation, either version 3 of the License, or
|
|
* (at your option) any later version.
|
|
*
|
|
* This program is distributed in the hope that it will be useful,
|
|
* but WITHOUT ANY WARRANTY; without even the implied warranty of
|
|
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
|
|
* GNU General Public License for more details.
|
|
*
|
|
* You should have received a copy of the GNU General Public License
|
|
* along with this program. If not, see <https://www.gnu.org/licenses/>.
|
|
*/
|
|
#include <MNIST.h>
|
|
#include <blt/fs/loader.h>
|
|
#include <blt/std/memory.h>
|
|
#include <blt/std/memory_util.h>
|
|
#include <variant>
|
|
#include <filesystem>
|
|
#include <iomanip>
|
|
#include <blt/iterator/iterator.h>
|
|
#include <blt/parse/argparse.h>
|
|
#include <blt/std/time.h>
|
|
#include <dlib/dnn.h>
|
|
#include <dlib/data_io.h>
|
|
#include <csignal>
|
|
|
|
namespace fp
|
|
{
|
|
constexpr blt::i64 batch_size = 256;
|
|
|
|
std::string binary_directory;
|
|
std::string python_dual_stacked_graph_program;
|
|
std::atomic_bool break_flag = false;
|
|
std::atomic_bool stop_flag = false;
|
|
std::atomic_bool learn_flag = false;
|
|
std::atomic_int64_t last_epoch = -1;
|
|
|
|
void run_python_line_graph(const std::string& title, const std::string& output_file, const std::string& csv1, const std::string& csv2,
|
|
const blt::size_t pos_forward, const blt::size_t pos_deep)
|
|
{
|
|
const auto command = "python3 '" + python_dual_stacked_graph_program + "' '" + title + "' '" + output_file + "' '" + csv1 + "' '" + csv2 + "' "
|
|
+ std::to_string(pos_forward) + " " + std::to_string(pos_deep);
|
|
BLT_TRACE("Running %s", command.c_str());
|
|
std::system(command.c_str());
|
|
}
|
|
|
|
class idx_file_t
|
|
{
|
|
template <typename T>
|
|
using mk_v = std::vector<T>;
|
|
using vec_t = std::variant<mk_v<blt::u8>, mk_v<blt::i8>, mk_v<blt::u16>, mk_v<blt::u32>, mk_v<blt::f32>, mk_v<blt::f64>>;
|
|
|
|
public:
|
|
explicit idx_file_t(const std::string& path)
|
|
{
|
|
std::ifstream file{path, std::ios::in | std::ios::binary};
|
|
|
|
using char_type = std::ifstream::char_type;
|
|
char_type magic_arr[4];
|
|
file.read(magic_arr, 4);
|
|
BLT_ASSERT(magic_arr[0] == 0 && magic_arr[1] == 0);
|
|
|
|
blt::u8 dims = magic_arr[3];
|
|
blt::size_t total_size = 1;
|
|
|
|
for (blt::i32 i = 0; i < dims; i++)
|
|
{
|
|
char_type dim_arr[4];
|
|
file.read(dim_arr, 4);
|
|
blt::u32 dim;
|
|
blt::mem::fromBytes(dim_arr, dim);
|
|
dimensions.push_back(dim);
|
|
total_size *= dim;
|
|
}
|
|
|
|
switch (magic_arr[2])
|
|
{
|
|
// unsigned char
|
|
case 0x08:
|
|
data = mk_v<blt::u8>{};
|
|
read_data<blt::u8>(file, total_size);
|
|
break;
|
|
// signed char
|
|
case 0x09:
|
|
data = mk_v<blt::i8>{};
|
|
read_data<blt::i8>(file, total_size);
|
|
break;
|
|
// short
|
|
case 0x0B:
|
|
data = mk_v<blt::u16>{};
|
|
read_data<blt::u16>(file, total_size);
|
|
reverse_data<blt::u16>();
|
|
break;
|
|
// int
|
|
case 0x0C:
|
|
data = mk_v<blt::u32>{};
|
|
read_data<blt::u32>(file, total_size);
|
|
reverse_data<blt::u32>();
|
|
break;
|
|
// float
|
|
case 0x0D:
|
|
data = mk_v<blt::f32>{};
|
|
read_data<blt::f32>(file, total_size);
|
|
reverse_data<blt::f32>();
|
|
break;
|
|
// double
|
|
case 0x0E:
|
|
data = mk_v<blt::f64>{};
|
|
read_data<blt::f64>(file, total_size);
|
|
reverse_data<blt::f64>();
|
|
break;
|
|
default:
|
|
BLT_ERROR("Unspported idx file type!");
|
|
}
|
|
if (file.eof())
|
|
{
|
|
BLT_ERROR("EOF reached. It's unlikely your file was read correctly!");
|
|
}
|
|
}
|
|
|
|
template <typename T>
|
|
[[nodiscard]] const std::vector<T>& get_data_as() const
|
|
{
|
|
return std::get<mk_v<T>>(data);
|
|
}
|
|
|
|
template <typename T>
|
|
std::vector<blt::span<T>> get_as_spans() const
|
|
{
|
|
std::vector<blt::span<T>> spans;
|
|
|
|
blt::size_t total_size = data_size(1);
|
|
|
|
for (blt::size_t i = 0; i < dimensions[0]; i++)
|
|
{
|
|
auto& array = std::get<mk_v<T>>(data);
|
|
spans.push_back({&array[i * total_size], total_size});
|
|
}
|
|
|
|
return spans;
|
|
}
|
|
|
|
[[nodiscard]] const std::vector<blt::u32>& get_dimensions() const
|
|
{
|
|
return dimensions;
|
|
}
|
|
|
|
[[nodiscard]] blt::size_t data_size(const blt::size_t starting_dimension = 0) const
|
|
{
|
|
blt::size_t total_size = 1;
|
|
for (const auto d : blt::iterate(dimensions).skip(starting_dimension))
|
|
total_size *= d;
|
|
return total_size;
|
|
}
|
|
|
|
private:
|
|
template <typename T>
|
|
void read_data(std::ifstream& file, blt::size_t total_size)
|
|
{
|
|
auto& array = std::get<mk_v<T>>(data);
|
|
array.resize(total_size);
|
|
file.read(reinterpret_cast<char*>(array.data()), static_cast<std::streamsize>(total_size) * sizeof(T));
|
|
}
|
|
|
|
template <typename T>
|
|
void reverse_data()
|
|
{
|
|
auto& array = std::get<mk_v<T>>(data);
|
|
for (auto& v : array)
|
|
blt::mem::reverse(v);
|
|
}
|
|
|
|
std::vector<blt::u32> dimensions;
|
|
vec_t data;
|
|
};
|
|
|
|
class image_t
|
|
{
|
|
public:
|
|
static constexpr blt::u32 target_size = 10;
|
|
using data_iterator = std::vector<dlib::matrix<blt::u8>>::const_iterator;
|
|
using label_iterator = std::vector<blt::u64>::const_iterator;
|
|
|
|
image_t(const idx_file_t& image_data, const idx_file_t& label_data): samples(image_data.get_dimensions()[0]),
|
|
input_size(image_data.data_size(1))
|
|
{
|
|
BLT_ASSERT_MSG(samples == label_data.get_dimensions()[0],
|
|
("Mismatch in data sample sizes! " + std::to_string(samples) + " vs " + std::to_string(label_data.get_dimensions()[0])).
|
|
c_str());
|
|
auto& image_array = image_data.get_data_as<blt::u8>();
|
|
auto& label_array = label_data.get_data_as<blt::u8>();
|
|
|
|
for (const auto label : label_array)
|
|
image_labels.push_back(label);
|
|
|
|
const auto row_length = image_data.get_dimensions()[2];
|
|
const auto number_of_rows = image_data.get_dimensions()[1];
|
|
|
|
for (blt::u32 i = 0; i < samples; i++)
|
|
{
|
|
dlib::matrix<blt::u8> mat(number_of_rows, row_length);
|
|
for (blt::u32 y = 0; y < number_of_rows; y++)
|
|
{
|
|
for (blt::u32 x = 0; x < row_length; x++)
|
|
{
|
|
mat(x, y) = image_array[i * input_size + y * row_length + x];
|
|
}
|
|
}
|
|
data.push_back(mat);
|
|
}
|
|
}
|
|
|
|
[[nodiscard]] const std::vector<dlib::matrix<blt::u8>>& get_image_data() const
|
|
{
|
|
return data;
|
|
}
|
|
|
|
[[nodiscard]] const std::vector<blt::u64>& get_image_labels() const
|
|
{
|
|
return image_labels;
|
|
}
|
|
|
|
private:
|
|
blt::u32 samples;
|
|
blt::u32 input_size;
|
|
std::vector<dlib::matrix<blt::u8>> data;
|
|
std::vector<blt::u64> image_labels;
|
|
};
|
|
|
|
struct batch_stats_t
|
|
{
|
|
blt::u64 hits = 0;
|
|
blt::u64 misses = 0;
|
|
|
|
friend std::ofstream& operator<<(std::ofstream& file, const batch_stats_t& stats)
|
|
{
|
|
file << stats.hits << ',' << stats.misses;
|
|
return file;
|
|
}
|
|
|
|
friend std::ifstream& operator>>(std::ifstream& file, batch_stats_t& stats)
|
|
{
|
|
file >> stats.hits;
|
|
file.ignore();
|
|
file >> stats.misses;
|
|
return file;
|
|
}
|
|
|
|
batch_stats_t& operator+=(const batch_stats_t& stats)
|
|
{
|
|
hits += stats.hits;
|
|
misses += stats.misses;
|
|
return *this;
|
|
}
|
|
|
|
batch_stats_t& operator/=(const blt::u64 divisor)
|
|
{
|
|
hits /= divisor;
|
|
misses /= divisor;
|
|
return *this;
|
|
}
|
|
};
|
|
|
|
struct epoch_stats_t
|
|
{
|
|
batch_stats_t test_results{};
|
|
double average_loss = 0;
|
|
double learn_rate = 0;
|
|
|
|
friend std::ofstream& operator<<(std::ofstream& file, const epoch_stats_t& stats)
|
|
{
|
|
file << stats.test_results << ',' << stats.average_loss << ',' << stats.learn_rate;
|
|
return file;
|
|
}
|
|
|
|
friend std::ifstream& operator>>(std::ifstream& file, epoch_stats_t& stats)
|
|
{
|
|
file >> stats.test_results;
|
|
file.ignore();
|
|
file >> stats.average_loss;
|
|
file.ignore();
|
|
file >> stats.learn_rate;
|
|
return file;
|
|
}
|
|
|
|
epoch_stats_t& operator+=(const epoch_stats_t& stats)
|
|
{
|
|
test_results += stats.test_results;
|
|
average_loss += stats.average_loss;
|
|
learn_rate += stats.learn_rate;
|
|
return *this;
|
|
}
|
|
|
|
epoch_stats_t& operator/=(const blt::u64 divisor)
|
|
{
|
|
test_results /= divisor;
|
|
average_loss /= static_cast<double>(divisor);
|
|
learn_rate /= static_cast<double>(divisor);
|
|
return *this;
|
|
}
|
|
};
|
|
|
|
struct network_stats_t
|
|
{
|
|
std::vector<epoch_stats_t> epoch_stats;
|
|
|
|
friend std::ofstream& operator<<(std::ofstream& file, const network_stats_t& stats)
|
|
{
|
|
file << stats.epoch_stats.size();
|
|
file << '\n';
|
|
for (const auto& v : stats.epoch_stats)
|
|
file << v << "\n";
|
|
return file;
|
|
}
|
|
|
|
friend std::ifstream& operator>>(std::ifstream& file, network_stats_t& stats)
|
|
{
|
|
blt::size_t size;
|
|
file >> size;
|
|
file.ignore();
|
|
for (blt::size_t i = 0; i < size; i++)
|
|
{
|
|
stats.epoch_stats.emplace_back();
|
|
file >> stats.epoch_stats.back();
|
|
file.ignore();
|
|
}
|
|
return file;
|
|
}
|
|
};
|
|
|
|
struct network_average_stats_t
|
|
{
|
|
std::vector<network_stats_t> run_stats;
|
|
|
|
network_average_stats_t& operator+=(const network_stats_t& stats)
|
|
{
|
|
run_stats.push_back(stats);
|
|
return *this;
|
|
}
|
|
|
|
[[nodiscard]] blt::size_t average_size() const
|
|
{
|
|
blt::size_t acc = 0;
|
|
for (const auto& [epoch_stats] : run_stats)
|
|
acc += epoch_stats.size();
|
|
return acc;
|
|
}
|
|
|
|
[[nodiscard]] network_stats_t average_stats() const
|
|
{
|
|
network_stats_t stats;
|
|
for (const auto& [epoch_stats] : run_stats)
|
|
{
|
|
if (stats.epoch_stats.size() < epoch_stats.size())
|
|
stats.epoch_stats.resize(epoch_stats.size());
|
|
for (const auto& [i, v] : blt::enumerate(epoch_stats))
|
|
{
|
|
stats.epoch_stats[i] += v;
|
|
}
|
|
}
|
|
for (auto& v : stats.epoch_stats)
|
|
v /= run_stats.size();
|
|
return stats;
|
|
}
|
|
|
|
friend std::ofstream& operator<<(std::ofstream& file, const network_average_stats_t& stats)
|
|
{
|
|
file << stats.run_stats.size();
|
|
file << '\n';
|
|
for (const auto& v : stats.run_stats)
|
|
file << v << "---\n";
|
|
return file;
|
|
}
|
|
|
|
friend std::ifstream& operator>>(std::ifstream& file, network_average_stats_t& stats)
|
|
{
|
|
blt::size_t size;
|
|
file >> size;
|
|
file.ignore();
|
|
for (blt::size_t i = 0; i < size; i++)
|
|
{
|
|
stats.run_stats.emplace_back();
|
|
file >> stats.run_stats.back();
|
|
file.ignore(4);
|
|
}
|
|
return file;
|
|
}
|
|
};
|
|
|
|
template <blt::i64 batch_size = batch_size, typename NetworkType>
|
|
batch_stats_t test_batch(NetworkType& network, image_t::data_iterator begin, const image_t::data_iterator end, image_t::label_iterator lbegin)
|
|
{
|
|
batch_stats_t stats{};
|
|
|
|
std::array<image_t::label_iterator::value_type, batch_size> output_labels{};
|
|
|
|
auto amount_remaining = std::distance(begin, end);
|
|
|
|
while (amount_remaining != 0)
|
|
{
|
|
const auto batch = std::min(amount_remaining, batch_size);
|
|
network(begin, begin + batch, output_labels.begin());
|
|
|
|
for (auto [predicted, expected] : blt::iterate(output_labels.begin(), output_labels.begin() + batch).zip(lbegin, lbegin + batch))
|
|
{
|
|
if (predicted == expected)
|
|
++stats.hits;
|
|
else
|
|
++stats.misses;
|
|
}
|
|
|
|
begin += batch;
|
|
lbegin += batch;
|
|
amount_remaining -= batch;
|
|
}
|
|
|
|
return stats;
|
|
}
|
|
|
|
template <typename NetworkType>
|
|
batch_stats_t test_network(NetworkType& network)
|
|
{
|
|
const idx_file_t test_images{binary_directory + "../problems/mnist/t10k-images.idx3-ubyte"};
|
|
const idx_file_t test_labels{binary_directory + "../problems/mnist/t10k-labels.idx1-ubyte"};
|
|
|
|
const image_t test_image{test_images, test_labels};
|
|
|
|
auto test_results = test_batch(network, test_image.get_image_data().begin(), test_image.get_image_data().end(),
|
|
test_image.get_image_labels().begin());
|
|
|
|
BLT_DEBUG("Testing hits: %lu", test_results.hits);
|
|
BLT_DEBUG("Testing misses: %lu", test_results.misses);
|
|
BLT_DEBUG("Testing accuracy: %lf", test_results.hits / static_cast<double>(test_results.hits + test_results.misses));
|
|
|
|
return test_results;
|
|
}
|
|
|
|
template <typename NetworkType>
|
|
network_stats_t train_network(const std::string& ident, NetworkType& network)
|
|
{
|
|
const idx_file_t train_images{binary_directory + "../problems/mnist/train-images.idx3-ubyte"};
|
|
const idx_file_t train_labels{binary_directory + "../problems/mnist/train-labels.idx1-ubyte"};
|
|
|
|
const image_t train_image{train_images, train_labels};
|
|
|
|
network_stats_t stats;
|
|
|
|
dlib::dnn_trainer trainer(network);
|
|
trainer.set_learning_rate(0.01);
|
|
trainer.set_min_learning_rate(0.00001);
|
|
trainer.set_mini_batch_size(batch_size);
|
|
trainer.set_max_num_epochs(100);
|
|
trainer.set_iterations_without_progress_threshold(2000);
|
|
trainer.be_verbose();
|
|
|
|
trainer.set_synchronization_file("mnist_sync_" + ident, std::chrono::seconds(20));
|
|
|
|
blt::size_t epochs = 0;
|
|
if (last_epoch > 0)
|
|
epochs = static_cast<blt::size_t>(last_epoch);
|
|
blt::ptrdiff_t epoch_pos = 0;
|
|
for (; epochs < trainer.get_max_num_epochs() && trainer.get_learning_rate() >= trainer.get_min_learning_rate(); epochs++)
|
|
{
|
|
auto& data = train_image.get_image_data();
|
|
auto& labels = train_image.get_image_labels();
|
|
for (; epoch_pos < data.size() && trainer.get_learning_rate() >= trainer.get_min_learning_rate(); epoch_pos += trainer.
|
|
get_mini_batch_size())
|
|
{
|
|
auto begin = epoch_pos;
|
|
auto end = std::min(epoch_pos + trainer.get_mini_batch_size(), data.size());
|
|
|
|
if (end - begin <= 0)
|
|
break;
|
|
|
|
if (learn_flag)
|
|
trainer.set_learning_rate(trainer.get_learning_rate() / 10);
|
|
|
|
trainer.train_one_step(train_image.get_image_data().begin() + begin,
|
|
data.begin() + end, labels.begin() + begin);
|
|
}
|
|
epoch_pos = 0;
|
|
BLT_TRACE("Trained an epoch (%ld/%ld) learn rate %lf average loss %lf", epochs, trainer.get_max_num_epochs(),
|
|
trainer.get_learning_rate(), trainer.get_average_loss());
|
|
|
|
// sync and test
|
|
trainer.get_net(dlib::force_flush_to_disk::no);
|
|
network.clean();
|
|
|
|
epoch_stats_t epoch_stats{};
|
|
epoch_stats.test_results = test_batch(network, train_image.get_image_data().begin(), train_image.get_image_data().end(),
|
|
train_image.get_image_labels().begin());
|
|
epoch_stats.average_loss = trainer.get_average_loss();
|
|
epoch_stats.learn_rate = trainer.get_learning_rate();
|
|
|
|
BLT_TRACE("\t\tHits: %lu\tMisses: %lu\tAccuracy: %lf", epoch_stats.test_results.hits, epoch_stats.test_results.misses,
|
|
epoch_stats.test_results.hits / static_cast<double>(epoch_stats.test_results.hits + epoch_stats.test_results.misses));
|
|
|
|
stats.epoch_stats.push_back(epoch_stats);
|
|
network.clean();
|
|
if (break_flag)
|
|
{
|
|
break_flag = false;
|
|
last_epoch = epochs;
|
|
break;
|
|
}
|
|
// dlib::serialize("mnist_network_" + ident + ".dat") << network;
|
|
}
|
|
|
|
BLT_INFO("Finished Training");
|
|
|
|
// sync
|
|
trainer.get_net();
|
|
network.clean();
|
|
|
|
// trainer.train(train_image.get_image_data(), train_image.get_image_labels());
|
|
dlib::serialize("mnist_network_" + ident + ".dat") << network;
|
|
|
|
auto test_results = test_batch(network, train_image.get_image_data().begin(), train_image.get_image_data().end(),
|
|
train_image.get_image_labels().begin());
|
|
|
|
BLT_DEBUG("Training hits: %lu", test_results.hits);
|
|
BLT_DEBUG("Training misses: %lu", test_results.misses);
|
|
BLT_DEBUG("Training accuracy: %lf", test_results.hits / static_cast<double>(test_results.hits + test_results.misses));
|
|
|
|
return stats;
|
|
}
|
|
|
|
template <typename NetworkType>
|
|
NetworkType load_network(const std::string& ident)
|
|
{
|
|
NetworkType network{};
|
|
dlib::deserialize("mnist_network_" + ident + ".dat") >> network;
|
|
return network;
|
|
}
|
|
|
|
template <typename NetworkType>
|
|
std::pair<network_average_stats_t, batch_stats_t> run_network_tests(std::string path, const std::string& ident, const blt::i32 runs,
|
|
const bool restore)
|
|
{
|
|
path += ("/" + ident + "/");
|
|
std::filesystem::create_directories(path);
|
|
std::filesystem::current_path(path);
|
|
|
|
network_average_stats_t stats{};
|
|
std::vector<batch_stats_t> test_stats;
|
|
|
|
blt::i32 i = 0;
|
|
if (std::filesystem::exists(path + "/state.bin"))
|
|
{
|
|
std::ifstream state{path + "/state.bin", std::ios::binary | std::ios::in};
|
|
if (!state.is_open())
|
|
{
|
|
BLT_ERROR("Failed to open state file!");
|
|
std::exit(-1);
|
|
}
|
|
|
|
state >> i;
|
|
state.ignore();
|
|
blt::i64 load_epoch = 0;
|
|
state >> load_epoch;
|
|
state.ignore();
|
|
last_epoch = load_epoch;
|
|
state >> stats;
|
|
state.ignore();
|
|
blt::size_t test_stats_size = 0;
|
|
state >> test_stats_size;
|
|
state.ignore();
|
|
for (blt::size_t _ = 0; _ < test_stats_size; _++)
|
|
{
|
|
test_stats.emplace_back();
|
|
state >> test_stats.back();
|
|
state.ignore();
|
|
}
|
|
}
|
|
|
|
blt::i64 last_epoch_save = -1;
|
|
for (; i < runs; i++)
|
|
{
|
|
if (stop_flag)
|
|
{
|
|
BLT_TRACE("Stopping!");
|
|
break;
|
|
}
|
|
BLT_TRACE("Starting run %d", i);
|
|
auto local_ident = ident + std::to_string(i);
|
|
NetworkType network{};
|
|
if (restore)
|
|
try
|
|
{
|
|
network = load_network<NetworkType>(local_ident);
|
|
}
|
|
catch (dlib::serialization_error&)
|
|
{
|
|
stats += train_network(local_ident, network);
|
|
}
|
|
else
|
|
stats += train_network(local_ident, network);
|
|
last_epoch_save = last_epoch;
|
|
last_epoch = -1;
|
|
test_stats.push_back(test_network(network));
|
|
}
|
|
|
|
batch_stats_t average;
|
|
for (const auto& v : test_stats)
|
|
average += v;
|
|
average /= runs;
|
|
|
|
std::ofstream state{path + "/state.bin", std::ios::binary | std::ios::out};
|
|
if (!state.is_open())
|
|
{
|
|
BLT_ERROR("Failed to open state file!");
|
|
std::exit(-1);
|
|
}
|
|
|
|
state << i;
|
|
state << '\n';
|
|
state << last_epoch_save;
|
|
state << '\n';
|
|
state << stats;
|
|
state << '\n';
|
|
state << test_stats.size();
|
|
state << '\n';
|
|
for (const auto& v : test_stats)
|
|
{
|
|
state << v;
|
|
state << '\n';
|
|
}
|
|
|
|
return {stats, average};
|
|
}
|
|
|
|
auto run_deep_learning_tests(const std::string& path, const blt::i32 runs, const bool restore)
|
|
{
|
|
using namespace dlib;
|
|
using net_type_dl = loss_multiclass_log<
|
|
fc<10,
|
|
relu<fc<84,
|
|
relu<fc<120,
|
|
max_pool<2, 2, 2, 2, relu<con<16, 5, 5, 1, 1,
|
|
max_pool<2, 2, 2, 2, relu<con<6, 5, 5, 1, 1,
|
|
input<matrix<blt::u8>>>>>>>>>>>>>>;
|
|
BLT_TRACE("Running deep learning tests");
|
|
return run_network_tests<net_type_dl>(path, "deep_learning", runs, restore);
|
|
}
|
|
|
|
auto run_feed_forward_tests(const std::string& path, const blt::i32 runs, const bool restore)
|
|
{
|
|
using namespace dlib;
|
|
|
|
using net_type_ff = loss_multiclass_log<
|
|
fc<10,
|
|
relu<fc<84,
|
|
relu<fc<120,
|
|
input<matrix<blt::u8>>>>>>>>;
|
|
|
|
BLT_TRACE("Running feed forward tests");
|
|
return run_network_tests<net_type_ff>(path, "feed_forward", runs, restore);
|
|
}
|
|
|
|
void run_mnist(const int argc, const char** argv)
|
|
{
|
|
binary_directory = std::filesystem::current_path();
|
|
blt::size_t pos = 0;
|
|
if (!blt::string::ends_with(binary_directory, '/'))
|
|
{
|
|
pos = binary_directory.find_last_of('/');
|
|
binary_directory += '/';
|
|
}
|
|
else
|
|
pos = binary_directory.substr(0, binary_directory.size() - 1).find_last_of('/');
|
|
python_dual_stacked_graph_program = binary_directory.substr(0, pos) + "/graph.py";
|
|
BLT_DEBUG(binary_directory);
|
|
BLT_DEBUG(python_dual_stacked_graph_program);
|
|
BLT_DEBUG("Running with batch size %d", batch_size);
|
|
|
|
BLT_DEBUG("Installing Signal Handlers");
|
|
if (std::signal(SIGINT, [](int)
|
|
{
|
|
BLT_INFO("Stopping current training");
|
|
break_flag = true;
|
|
}) == SIG_ERR)
|
|
{
|
|
BLT_ERROR("Failed to replace SIGINT");
|
|
}
|
|
if (std::signal(SIGQUIT, [](int)
|
|
{
|
|
BLT_INFO("Exiting Program");
|
|
stop_flag = true;
|
|
break_flag = true;
|
|
}) == SIG_ERR)
|
|
{
|
|
BLT_ERROR("Failed to replace SIGQUIT");
|
|
}
|
|
if (std::signal(SIGUSR1, [](int)
|
|
{
|
|
BLT_INFO("Decreasing Learn Rate for current training");
|
|
learn_flag = true;
|
|
}) == SIG_ERR)
|
|
{
|
|
BLT_ERROR("Failed to replace SIGUSR1");
|
|
}
|
|
if (std::signal(SIGUSR2, [](int)
|
|
{
|
|
BLT_INFO("Exiting Program");
|
|
stop_flag = true;
|
|
break_flag = true;
|
|
}) == SIG_ERR)
|
|
{
|
|
BLT_ERROR("Failed to replace SIGUSR2");
|
|
}
|
|
|
|
using namespace dlib;
|
|
|
|
blt::arg_parse parser{};
|
|
parser.addArgument(
|
|
blt::arg_builder{"-r", "--restore"}.setAction(blt::arg_action_t::STORE_TRUE).setDefault(false).setHelp(
|
|
"Restores from last save").build());
|
|
parser.addArgument(blt::arg_builder{"-t", "--runs"}.setHelp("Number of runs to perform [default: 10]").setDefault("10").build());
|
|
parser.addArgument(
|
|
blt::arg_builder{"-p", "--python"}.setHelp("Only run the python scripts").setAction(blt::arg_action_t::STORE_TRUE).setDefault(false).
|
|
build());
|
|
parser.addArgument(
|
|
blt::arg_builder{"network"}.setDefault(std::to_string(blt::system::getCurrentTimeMilliseconds())).setHelp("location of network files").
|
|
build());
|
|
|
|
auto args = parser.parse_args(argc, argv);
|
|
|
|
const auto runs = std::stoi(args.get<std::string>("runs"));
|
|
const auto restore = args.get<bool>("restore");
|
|
auto path = binary_directory + args.get<std::string>("network");
|
|
|
|
auto [deep_stats, deep_tests] = run_deep_learning_tests(path, runs, restore);
|
|
auto [forward_stats, forward_tests] = run_feed_forward_tests(path, runs, restore);
|
|
|
|
auto average_forward_size = forward_stats.average_size();
|
|
auto average_deep_size = deep_stats.average_size();
|
|
|
|
{
|
|
std::ofstream test_results_f{path + "/test_results_table.txt"};
|
|
test_results_f << "\\begin{figure}" << std::endl;
|
|
test_results_f << "\t\\begin{tabular}{|c|c|c|c|}" << std::endl;
|
|
test_results_f << "\t\t\\hline" << std::endl;
|
|
test_results_f << "\t\tTest & Correct & Incorrect & Accuracy (\\%) \\\\" << std::endl;
|
|
test_results_f << "\t\t\\hline" << std::endl;
|
|
auto test_accuracy = forward_tests.hits / static_cast<double>(forward_tests.hits + forward_tests.misses) * 100;
|
|
test_results_f << "\t\tFeed-Forward & " << forward_tests.hits << " & " << forward_tests.misses << " & " << std::setprecision(2) <<
|
|
test_accuracy << "\\\\" << std::endl;
|
|
test_accuracy = deep_tests.hits / static_cast<double>(deep_tests.hits + deep_tests.misses) * 100;
|
|
test_results_f << "\t\tDeep Learning & " << deep_tests.hits << " & " << deep_tests.misses << " & " << std::setprecision(2) <<
|
|
test_accuracy << "\\\\" << std::endl;
|
|
test_results_f << "\t\\end{tabular}" << std::endl;
|
|
test_results_f << "\\end{figure}" << std::endl;
|
|
|
|
const auto [forward_epoch_stats] = forward_stats.average_stats();
|
|
std::ofstream train_forward{path + "/forward_train_results.csv"};
|
|
train_forward << "Epoch,Loss" << std::endl;
|
|
for (const auto& [i, v] : blt::enumerate(forward_epoch_stats))
|
|
train_forward << i << ',' << v.average_loss << std::endl;
|
|
|
|
const auto [deep_epoch_stats] = deep_stats.average_stats();
|
|
std::ofstream train_deep{path + "/deep_train_results.csv"};
|
|
train_deep << "Epoch,Loss" << std::endl;
|
|
for (const auto& [i, v] : blt::enumerate(deep_epoch_stats))
|
|
train_deep << i << ',' << v.average_loss << std::endl;
|
|
|
|
std::ofstream average_epochs{path + "/average_epochs.txt"};
|
|
average_epochs << average_forward_size << "," << average_deep_size << std::endl;
|
|
}
|
|
|
|
run_python_line_graph("Feed-Forward vs Deep Learning, Average Loss over Epochs", "epochs.png", path + "/forward_train_results.csv",
|
|
path + "/deep_train_results.csv", average_forward_size, average_deep_size);
|
|
|
|
// net_type_dl test_net;
|
|
// const auto stats = train_network("dl_nn", test_net);
|
|
// std::ofstream out_file{"dl_nn.csv"};
|
|
// out_file << stats;
|
|
|
|
// test_net = load_network<net_type_dl>("dl_nn");
|
|
|
|
// test_network(test_net);
|
|
}
|
|
}
|