# 🔥 Ember
Ember is a statistics and ML library for my personal use with C++ and Python. I mainly built it for educational purposes, but it's quite functional and can be used to train several datasets.
- [Installation](#installation)
- [Compilation](#compiling-the-aten-library)
- [Testing](#testing)
- [Repository Structure](#repository-structure)
- [Getting Started](#getting-started)
- [Ember Tensors and GradTensors](#ember-tensors-and-gradtensors)
- [Automatic Differentiation](#automatic-differentiation)
- [Linear Regression](#linear-regression)
- [Neural Nets](#multilayer-perceptrons)
- [Datasets](#datasets)
- [Models and Training](#models-and-training)
- [Monte Carlo Samplers](#monte-carlo-samplers)
Look [here](docs/progress.md) to see the methods it supports.
## Installation
### System Support
This library supports both x86_64/amd64 and arm64/aarch64. Check if your system is supported out of the box in the table below. The library requires very few dependencies, so as long as your machine supports a C++ compiler and python, you should be able to get it working by fiddling with the CMake and setuptools files.
|x86_64|Python 3.13|Python 3.12|Python 3.11|Python 3.10|Python 3.9|Python 3.8|Python 3.7|
|-|-|-|-|-|-|-|-|
|Ubuntu 24.04 | |✅| | | | | |
|Ubuntu 22.04 | |✅| | | | | |
|Ubuntu 20.04 | |✅| | | | | |
|ArchLinux 6.6.68 LTS | || | | | | |
|Debian 13 | || | | | | |
|Debian 12 | |✅| | | | | |
|Debian 11 | |✅| | | | | |
|Debian 10 | |❌| | | | | |
|LinuxMint 22 | || | | | | |
|LinuxMint 21 | || | | | | |
|MacOS 10.15 Catalina | || | | | | |
|MacOS 10.14 Mojave | || | | | | |
|MacOS 10.13 High Sierra | || | | | | |
|MacOS 10.12 Sierra | || | | | | |
|MacOS 10.11 El Capitan | || | | | | |
|MacOS 10.10 Yosemite | || | | | | |
|MacOS 10.9 Mavericks | || | | | | |
|MacOS 10.8 Mountain Lion | || | | | | |
|MacOS 10.7 Lion | || | | | | |
|Windows 11 | |✅| | | | | |
|Windows 10 | || | | | | |
|Windows 8 | || | | | | |
|Windows 7 | || | | | | |
|ARM64|Python 3.13|Python 3.12|Python 3.11|Python 3.10|Python 3.9|Python 3.8|Python 3.7|
|-|-----------|-----------|-----------|-----------|----------|----------|----------|
|Ubuntu 24.04 | |✅| | | | | |
|Ubuntu 22.04 | |✅| | | | | |
|Ubuntu 20.04 | |✅| | | | | |
|MacOS 15.x Sequoia | || | | | | |
|MacOS 14.x Sonoma | |✅| | | | | |
|MacOS 13.x Ventura | || | | | | |
|MacOS 12.x Monterey | || | | | | |
|MacOS 11.x Big Sur | || | | | | |
|Windows 12 | | | | | | | |
### Compiling the `aten` Library
Your machine will need system dependencies such as CMake, a C++ compiler, and pybind11. The library uses C++17. Preferably you will have git and conda installed already. For more specific instructions on installing these on your system, refer to the more detailed installation guide.
Git clone the repo, then pip install, which will run `setup.py`.
```
git clone git@github.com:mbahng/pyember.git
cd pyember
pip install .
```
This runs `cmake` on `aten/CMakeLists.txt`, which calls the following.
1. It always calls `aten/src/CMakeLists.txt` that compiles and links the source files in the C++ tensor library.
2. If `BUILD_PYTHON_BINDINGS=ON` (always on by default), it further calls `aten/bindings/CMakeLists.txt` to further generate a `.so` file that can be imported into `ember`.
3. If `BUILD_DEV=ON`, it calls `aten/test/CMakeLists.txt` to further compile the C++ unit testing suite.
If there are problems with building, you should check, in order,
1. Whether `build/` has been created. This is the first step in `setup.py`
2. Whether the compiled `main.cpp` and, if `BUILD_DEV=ON`, the C++ unit test files have been compiled, i.e. if `build/src/main` and `build/test/tests` executables exist.
3. Whether `build/*/aten.cpython-3**-darwin.so` exists (somewhere in the build directory, depending on the machine). The Makefile generated by `aten/bindings/CMakeLists.txt` will produce `build/*/aten.cpython-3**-darwin.so`.
4. The `setup()` function will immediately copy this `.so` file to `ember/aten.cpython-3**-darwin.so`. You should see a success message saying that it has been moved or an error. The `.so` file must live within `ember`, the actual library, since `ember/__init__.py` must access it within the same directory level.
### Testing and Development
The pip install comes with two more environment variable parameters. Note that the following command is whitespace-sensitive.
```
CMAKE_DEBUG=1 CMAKE_DEV=1 pip install .
```
1. Setting `CMAKE_DEBUG=1` compiles the `aten` library with debug mode (`-g`) on, which I use when using gdb/lldb on the compiled code.
2. Setting `CMAKE_DEV=1` compiles the C++ testing suite as well. If you want to do this, you will also need to install google-tests. A code snippet for Ubuntu and Debian is shown below.
```
sudo apt-get install libgtest-dev
cd /usr/src/gtest
cmake CMakeLists.txt
make
cp lib/*.a /usr/lib
rm -rf /var/lib/apt/lists/*
```
If you would like to run tests and/or develop the package yourself, you can run the script `./run_tests.sh all` (args `python` to run just python tests and `cpp` to run just C++ tests), which will
1. Run all C++ unit tests for `aten`, ensuring that all functions work correctly.
2. Run all Python unit tests for `ember`, ensuring that additional functions work correctly and that the C++ functions are bound correctly.
The stub (`.pyi`) files for `aten` are located in `ember/aten`.
## Repository Structure
I tried to model a lot of the structure from Pytorch and TinyGrad. Very briefly,
1. `aten/` contains the header and source files for the C++ low-level tensor library, such as basic operations and an autograd engine.
1. `aten/src` contains all the source files and definitions.
2. `aten/bindings` contains the pybindings.
3. `aten/test` contains all the C++ testing modules for aten.
2. `ember/` contains the actual library, supporting high level models, objectives, optimizers, dataloaders, and samplers.
1. `ember/aten` contains the stub files.
2. `ember/datasets` contains all preprocessing tools, such as datasets/loaders, standardizing, cross validation checks.
3. `ember/models` contains all machine learning models.
4. `ember/objectives` contain all loss functions and regularizers.
5. `ember/optimizers` contain all the optimizers/solvers, such as iterative (e.g. SGD), greedy (e.g. decision tree splitting), and one-shot (e.g. least-squares solution).
6. `ember/samplers` contain all samplers (e.g. MCMC, SGLD).
3. `docs/` contains detailed documentation about each function.
4. `examples/` are example python scripts on training models.
5. `tests/` are python testing modules for the `ember` library.
6. `docker/` contains docker images of all the operating systems and architectures I tested ember on. General workflows on setting up the environment can be found there for supported machines.
7. `setup.py` allows you to pip install this as a package.
8. `run_tests.sh` which is the main test running script.
For a more detailed explanation, look [here](docs/structure.md).
## Getting Started
### Ember Tensors and GradTensors
`ember.Tensor`s represent data and parameters, while `ember.GradTensor`s represent gradients. An advantage of this package is that rather than just supporting batch vector operations and matrix multiplications, we can also perform general contractions of rank $(N, M)$-tensors, a generalization of matrix multiplication. This allows us to represent and utilize the full power of higher order derivatives for arbitrary functions $f: \mathbb{R}^{\mathbf{M}} \rightarrow \mathbb{R}^{\mathbf{N}}$, where $\mathbf{M} = (M_1, \ldots, M_m)$ and $\mathbf{N} = (N_1, \ldots, N_m)$ are *vectors*, not just scalars, representing the dimension of each space.
Tensors are multidimensional arrays that can be initialized in a number of ways. GradTensors are initialized during the backpropagation method, but we can explicitly set them if desired.
```
import ember
a = ember.Tensor([2]) # scalar
b = ember.Tensor([1, 2, 3]) # vector
c = ember.Tensor([[1, 2], [3, 4]]) # 2D vector
d = ember.Tensor([[[1, 2]]]) # 3D vector
```
Say that you have a series of elementary operations on tensors.
```
a = ember.Tensor([2, -3])
h = a ** 2
b = ember.Tensor([3, 5])
c = b * h
d = ember.Tensor([10, 1])
e = c.dot(d)
f = ember.Tensor([-2])
g = f * e
```
### Automatic Differentiation
The C++ backend computes a directed acyclic graph (DAG) representing the operations done to compute `g`. You can then run `g.backprop()` to compute the gradients by applying the chain rule. This constructs the DAG and returns a topological sorting of its nodes. The gradients themselves, which are technically Jacobian matrices, are updated, with each mapping `x -> y` constructing a gradient tensor on `x` with value `dy/dx`. The gradients can be either accumulated by setting `backprop(intermediate=False)` so that the chain rule is not applied yet, or we can set `=True` to apply the chain rule to calculate the derivative of the tensor we called backprop on w.r.t. the rest of the tensors.
```
top_sort = g.backprop()
print(a.grad) # [[4.0, 0.0], [0.0, -6.0]]
print(h.grad) # [[3.0, 0.0], [0.0, 5.0]]
print(b.grad) # [[4.0, 0.0], [0.0, 9.0]]
print(c.grad) # [[10.0, 1.0]]
print(d.grad) # [[12.0, 45.0]]
print(e.grad) # [[-2.0]]
print(f.grad) # [[165.0]]
print(g.grad) # [[1.0]]
```
Finally, we can visualize this using the `networkx` package.
### Linear Regression
To perform linear regression, use the `LinearRegression` model.
```
import ember
ds = ember.datasets.LinearDataset(N=20, D=14)
dl = ember.datasets.Dataloader(ds, batch_size=2)
model = ember.models.LinearRegression(15)
mse = ember.objectives.MSELoss()
for epoch in range(500):
loss = None
for x, y in dl:
y_ = model.forward(x)
loss = mse(y, y_)
loss.backprop()
model.step(1e-5)
print(loss)
```
### K Nearest Neighbors
To do a simple K Nearest Neighbors regressor, use the following model. The forward method scans over the whole dataset, so we must input it to the model during instantiation. Note that we do not need a dataloader or a backpropagation method since we aren't iteratively updating gradients, though we want to show the loss.
```
import ember
from ember.models import KNearestRegressor
from ember.datasets import LinearDataset
ds = LinearDataset(N=20, D=3)
model = KNearestRegressor(dataset=ds, K=1)
mse = ember.objectives.MSELoss()
for k in range(1, 21): # hyperparameter tuning
model.K = k
print(f"{k} ===")
loss = 0
for i in range(len(ds)):
x, y = ds[i]
y_ = model.forward(x)
loss = loss + mse(y, y_)
print(loss)
```
### Multilayer Perceptrons
To instantiate a MLP, just call it from models. In here we make a 2-layer MLP with a dummy dataset. For now only SGD with batch size 1 is supported.
```
import ember
ds = ember.datasets.LinearDataset(N=20, D=14)
dl = ember.datasets.Dataloader(ds, batch_size=2)
model = ember.models.MultiLayerPerceptron(15, 10)
mse = ember.objectives.MSELoss()
for epoch in range(500):
loss = None
for x, y in dl:
y_ = model.forward(x)
loss = mse(y, y_)
loss.backprop()
model.step(1e-5)
print(loss)
```
Its outputs over 1 minute.
```
LOSS = 256733.64437981808
LOSS = 203239.08846901066
LOSS = 160223.4554735339
LOSS = 125704.33716141782
LOSS = 98074.96981384761
LOSS = 76026.19871949886
LOSS = 58491.92389906721
LOSS = 44604.493032865605
LOSS = 33658.23285350788
LOSS = 25079.638682869212
LOSS = 18403.01062298029
LOSS = 13250.54496118543
LOSS = 9316.069468116035
LOSS = 6351.758695807299
LOSS = 4157.286052245369
LOSS = 2570.96819208677
LOSS = 1462.5380952427417
LOSS = 727.2493587808174
LOSS = 281.0683664354656
LOSS = 56.75530418715159
```
### Datasets
### Models and Training
### Monte Carlo Samplers
## Contributing
To implement a new functionality in the `aten` library, you must
1. Add the class or function header in `aten/src/Tensor.h`
2. Add the implementation in the correct file (or create a new one) in `aten./*Tensor/*.cpp`. Make sure to update `aten/bindings/CMakeLists.txt` if needed.
3. Add its pybindings (if a public function that will be used in `ember`) in `aten/bindings/*bindings.cpp`. Make sure to update `aten/bindings/CMakeLists.txt` if needed.
4. Add relevant C++ tests in `aten/test/`.
5. Not necessary, but it's good to test it out on a personal script for a sanity check.
6. Add to the stub files in `ember/aten/*.pyi`.
7. Add Python tests in `test/`.
8. If everything passes, you can submit a pull request.
Raw data
{
"_id": null,
"home_page": null,
"name": "pyember",
"maintainer": null,
"docs_url": null,
"requires_python": ">=3.9",
"maintainer_email": null,
"keywords": "machine learning, autograd, deep learning, statistics",
"author": null,
"author_email": "Muchang Bahng <bahngmc@gmail.com>",
"download_url": null,
"platform": null,
"description": "# \ud83d\udd25 Ember\n\nEmber is a statistics and ML library for my personal use with C++ and Python. I mainly built it for educational purposes, but it's quite functional and can be used to train several datasets. \n\n- [Installation](#installation) \n - [Compilation](#compiling-the-aten-library)\n - [Testing](#testing) \n- [Repository Structure](#repository-structure) \n- [Getting Started](#getting-started)\n - [Ember Tensors and GradTensors](#ember-tensors-and-gradtensors)\n - [Automatic Differentiation](#automatic-differentiation)\n - [Linear Regression](#linear-regression)\n - [Neural Nets](#multilayer-perceptrons)\n - [Datasets](#datasets)\n - [Models and Training](#models-and-training)\n - [Monte Carlo Samplers](#monte-carlo-samplers)\n\nLook [here](docs/progress.md) to see the methods it supports. \n\n## Installation \n\n### System Support \n\nThis library supports both x86_64/amd64 and arm64/aarch64. Check if your system is supported out of the box in the table below. The library requires very few dependencies, so as long as your machine supports a C++ compiler and python, you should be able to get it working by fiddling with the CMake and setuptools files. \n\n|x86_64|Python 3.13|Python 3.12|Python 3.11|Python 3.10|Python 3.9|Python 3.8|Python 3.7|\n|-|-|-|-|-|-|-|-|\n|Ubuntu 24.04 | |\u2705| | | | | |\n|Ubuntu 22.04 | |\u2705| | | | | |\n|Ubuntu 20.04 | |\u2705| | | | | |\n|ArchLinux 6.6.68 LTS | || | | | | |\n|Debian 13 | || | | | | |\n|Debian 12 | |\u2705| | | | | |\n|Debian 11 | |\u2705| | | | | |\n|Debian 10 | |\u274c| | | | | |\n|LinuxMint 22 | || | | | | |\n|LinuxMint 21 | || | | | | |\n|MacOS 10.15 Catalina | || | | | | |\n|MacOS 10.14 Mojave | || | | | | |\n|MacOS 10.13 High Sierra | || | | | | |\n|MacOS 10.12 Sierra | || | | | | |\n|MacOS 10.11 El Capitan | || | | | | |\n|MacOS 10.10 Yosemite | || | | | | |\n|MacOS 10.9 Mavericks | || | | | | |\n|MacOS 10.8 Mountain Lion | || | | | | |\n|MacOS 10.7 Lion | || | | | | |\n|Windows 11 | |\u2705| | | | | |\n|Windows 10 | || | | | | |\n|Windows 8 | || | | | | |\n|Windows 7 | || | | | | |\n\n|ARM64|Python 3.13|Python 3.12|Python 3.11|Python 3.10|Python 3.9|Python 3.8|Python 3.7|\n|-|-----------|-----------|-----------|-----------|----------|----------|----------|\n|Ubuntu 24.04 | |\u2705| | | | | |\n|Ubuntu 22.04 | |\u2705| | | | | |\n|Ubuntu 20.04 | |\u2705| | | | | |\n|MacOS 15.x Sequoia | || | | | | |\n|MacOS 14.x Sonoma | |\u2705| | | | | |\n|MacOS 13.x Ventura | || | | | | |\n|MacOS 12.x Monterey | || | | | | |\n|MacOS 11.x Big Sur | || | | | | |\n|Windows 12 | | | | | | | |\n\n### Compiling the `aten` Library \n\nYour machine will need system dependencies such as CMake, a C++ compiler, and pybind11. The library uses C++17. Preferably you will have git and conda installed already. For more specific instructions on installing these on your system, refer to the more detailed installation guide. \n\nGit clone the repo, then pip install, which will run `setup.py`. \n\n```\ngit clone git@github.com:mbahng/pyember.git \ncd pyember \npip install .\n```\n\nThis runs `cmake` on `aten/CMakeLists.txt`, which calls the following. \n1. It always calls `aten/src/CMakeLists.txt` that compiles and links the source files in the C++ tensor library. \n2. If `BUILD_PYTHON_BINDINGS=ON` (always on by default), it further calls `aten/bindings/CMakeLists.txt` to further generate a `.so` file that can be imported into `ember`. \n3. If `BUILD_DEV=ON`, it calls `aten/test/CMakeLists.txt` to further compile the C++ unit testing suite. \n\nIf there are problems with building, you should check, in order, \n1. Whether `build/` has been created. This is the first step in `setup.py` \n2. Whether the compiled `main.cpp` and, if `BUILD_DEV=ON`, the C++ unit test files have been compiled, i.e. if `build/src/main` and `build/test/tests` executables exist. \n3. Whether `build/*/aten.cpython-3**-darwin.so` exists (somewhere in the build directory, depending on the machine). The Makefile generated by `aten/bindings/CMakeLists.txt` will produce `build/*/aten.cpython-3**-darwin.so`. \n4. The `setup()` function will immediately copy this `.so` file to `ember/aten.cpython-3**-darwin.so`. You should see a success message saying that it has been moved or an error. The `.so` file must live within `ember`, the actual library, since `ember/__init__.py` must access it within the same directory level. \n\n### Testing and Development\n\nThe pip install comes with two more environment variable parameters. Note that the following command is whitespace-sensitive. \n```\nCMAKE_DEBUG=1 CMAKE_DEV=1 pip install .\n```\n1. Setting `CMAKE_DEBUG=1` compiles the `aten` library with debug mode (`-g`) on, which I use when using gdb/lldb on the compiled code. \n2. Setting `CMAKE_DEV=1` compiles the C++ testing suite as well. If you want to do this, you will also need to install google-tests. A code snippet for Ubuntu and Debian is shown below. \n```\nsudo apt-get install libgtest-dev \ncd /usr/src/gtest \ncmake CMakeLists.txt \nmake \ncp lib/*.a /usr/lib \nrm -rf /var/lib/apt/lists/*\n```\n\nIf you would like to run tests and/or develop the package yourself, you can run the script `./run_tests.sh all` (args `python` to run just python tests and `cpp` to run just C++ tests), which will \n1. Run all C++ unit tests for `aten`, ensuring that all functions work correctly. \n2. Run all Python unit tests for `ember`, ensuring that additional functions work correctly and that the C++ functions are bound correctly. \n\nThe stub (`.pyi`) files for `aten` are located in `ember/aten`. \n\n## Repository Structure \n\n I tried to model a lot of the structure from Pytorch and TinyGrad. Very briefly, \n\n 1. `aten/` contains the header and source files for the C++ low-level tensor library, such as basic operations and an autograd engine. \n 1. `aten/src` contains all the source files and definitions. \n 2. `aten/bindings` contains the pybindings. \n 3. `aten/test` contains all the C++ testing modules for aten. \n 2. `ember/` contains the actual library, supporting high level models, objectives, optimizers, dataloaders, and samplers. \n 1. `ember/aten` contains the stub files. \n 2. `ember/datasets` contains all preprocessing tools, such as datasets/loaders, standardizing, cross validation checks. \n 3. `ember/models` contains all machine learning models. \n 4. `ember/objectives` contain all loss functions and regularizers. \n 5. `ember/optimizers` contain all the optimizers/solvers, such as iterative (e.g. SGD), greedy (e.g. decision tree splitting), and one-shot (e.g. least-squares solution). \n 6. `ember/samplers` contain all samplers (e.g. MCMC, SGLD). \n 3. `docs/` contains detailed documentation about each function. \n 4. `examples/` are example python scripts on training models. \n 5. `tests/` are python testing modules for the `ember` library. \n 6. `docker/` contains docker images of all the operating systems and architectures I tested ember on. General workflows on setting up the environment can be found there for supported machines. \n 7. `setup.py` allows you to pip install this as a package. \n 8. `run_tests.sh` which is the main test running script. \n\n For a more detailed explanation, look [here](docs/structure.md). \n\n## Getting Started \n\n### Ember Tensors and GradTensors\n\n`ember.Tensor`s represent data and parameters, while `ember.GradTensor`s represent gradients. An advantage of this package is that rather than just supporting batch vector operations and matrix multiplications, we can also perform general contractions of rank $(N, M)$-tensors, a generalization of matrix multiplication. This allows us to represent and utilize the full power of higher order derivatives for arbitrary functions $f: \\mathbb{R}^{\\mathbf{M}} \\rightarrow \\mathbb{R}^{\\mathbf{N}}$, where $\\mathbf{M} = (M_1, \\ldots, M_m)$ and $\\mathbf{N} = (N_1, \\ldots, N_m)$ are *vectors*, not just scalars, representing the dimension of each space. \n\nTensors are multidimensional arrays that can be initialized in a number of ways. GradTensors are initialized during the backpropagation method, but we can explicitly set them if desired. \n```\nimport ember \n\na = ember.Tensor([2]) # scalar\nb = ember.Tensor([1, 2, 3]) # vector \nc = ember.Tensor([[1, 2], [3, 4]]) # 2D vector \nd = ember.Tensor([[[1, 2]]]) # 3D vector\n```\nSay that you have a series of elementary operations on tensors. \n```\na = ember.Tensor([2, -3]) \nh = a ** 2\nb = ember.Tensor([3, 5])\n\nc = b * h\n\nd = ember.Tensor([10, 1])\ne = c.dot(d)\n\nf = ember.Tensor([-2])\n\ng = f * e\n```\n\n### Automatic Differentiation\n\nThe C++ backend computes a directed acyclic graph (DAG) representing the operations done to compute `g`. You can then run `g.backprop()` to compute the gradients by applying the chain rule. This constructs the DAG and returns a topological sorting of its nodes. The gradients themselves, which are technically Jacobian matrices, are updated, with each mapping `x -> y` constructing a gradient tensor on `x` with value `dy/dx`. The gradients can be either accumulated by setting `backprop(intermediate=False)` so that the chain rule is not applied yet, or we can set `=True` to apply the chain rule to calculate the derivative of the tensor we called backprop on w.r.t. the rest of the tensors. \n\n```\ntop_sort = g.backprop()\nprint(a.grad) # [[4.0, 0.0], [0.0, -6.0]]\nprint(h.grad) # [[3.0, 0.0], [0.0, 5.0]]\nprint(b.grad) # [[4.0, 0.0], [0.0, 9.0]]\nprint(c.grad) # [[10.0, 1.0]]\nprint(d.grad) # [[12.0, 45.0]]\nprint(e.grad) # [[-2.0]]\nprint(f.grad) # [[165.0]]\nprint(g.grad) # [[1.0]]\n```\n\n\nFinally, we can visualize this using the `networkx` package. \n\n\n\n### Linear Regression \n\nTo perform linear regression, use the `LinearRegression` model. \n```\nimport ember \n\nds = ember.datasets.LinearDataset(N=20, D=14)\ndl = ember.datasets.Dataloader(ds, batch_size=2)\nmodel = ember.models.LinearRegression(15) \nmse = ember.objectives.MSELoss()\n\nfor epoch in range(500): \n loss = None\n for x, y in dl: \n y_ = model.forward(x) \n loss = mse(y, y_)\n loss.backprop()\n model.step(1e-5) \n\n print(loss)\n``` \n\n### K Nearest Neighbors \n\nTo do a simple K Nearest Neighbors regressor, use the following model. The forward method scans over the whole dataset, so we must input it to the model during instantiation. Note that we do not need a dataloader or a backpropagation method since we aren't iteratively updating gradients, though we want to show the loss. \n\n```\nimport ember\nfrom ember.models import KNearestRegressor\nfrom ember.datasets import LinearDataset\n\nds = LinearDataset(N=20, D=3)\nmodel = KNearestRegressor(dataset=ds, K=1)\nmse = ember.objectives.MSELoss() \n\nfor k in range(1, 21): # hyperparameter tuning\n model.K = k\n print(f\"{k} ===\") \n loss = 0\n for i in range(len(ds)): \n x, y = ds[i] \n y_ = model.forward(x) \n loss = loss + mse(y, y_) \n\n print(loss)\n```\n\n### Multilayer Perceptrons \n\nTo instantiate a MLP, just call it from models. In here we make a 2-layer MLP with a dummy dataset. For now only SGD with batch size 1 is supported. \n```\nimport ember \n\nds = ember.datasets.LinearDataset(N=20, D=14)\ndl = ember.datasets.Dataloader(ds, batch_size=2)\nmodel = ember.models.MultiLayerPerceptron(15, 10) \nmse = ember.objectives.MSELoss()\n\nfor epoch in range(500): \n loss = None\n for x, y in dl: \n y_ = model.forward(x) \n loss = mse(y, y_)\n loss.backprop() \n model.step(1e-5)\n\n print(loss)\n```\nIts outputs over 1 minute. \n```\nLOSS = 256733.64437981808\nLOSS = 203239.08846901066\nLOSS = 160223.4554735339\nLOSS = 125704.33716141782\nLOSS = 98074.96981384761\nLOSS = 76026.19871949886\nLOSS = 58491.92389906721\nLOSS = 44604.493032865605\nLOSS = 33658.23285350788\nLOSS = 25079.638682869212\nLOSS = 18403.01062298029\nLOSS = 13250.54496118543\nLOSS = 9316.069468116035\nLOSS = 6351.758695807299\nLOSS = 4157.286052245369\nLOSS = 2570.96819208677\nLOSS = 1462.5380952427417\nLOSS = 727.2493587808174\nLOSS = 281.0683664354656\nLOSS = 56.75530418715159\n```\n\n### Datasets\n\n### Models and Training\n\n### Monte Carlo Samplers\n\n## Contributing \n\nTo implement a new functionality in the `aten` library, you must \n1. Add the class or function header in `aten/src/Tensor.h` \n2. Add the implementation in the correct file (or create a new one) in `aten./*Tensor/*.cpp`. Make sure to update `aten/bindings/CMakeLists.txt` if needed.\n3. Add its pybindings (if a public function that will be used in `ember`) in `aten/bindings/*bindings.cpp`. Make sure to update `aten/bindings/CMakeLists.txt` if needed. \n4. Add relevant C++ tests in `aten/test/`. \n5. Not necessary, but it's good to test it out on a personal script for a sanity check. \n6. Add to the stub files in `ember/aten/*.pyi`. \n7. Add Python tests in `test/`. \n8. If everything passes, you can submit a pull request. \n\n",
"bugtrack_url": null,
"license": null,
"summary": "A statistics and machine learning package.",
"version": "0.0.16",
"project_urls": {
"Homepage": "https://github.com/mbahng/pyember"
},
"split_keywords": [
"machine learning",
" autograd",
" deep learning",
" statistics"
],
"urls": [
{
"comment_text": "",
"digests": {
"blake2b_256": "e89ad0c9abe9bb464b2703394d6fd0f4af74e58ccc5db26fbdcc15b1e69d4c4f",
"md5": "3f57a707b22d153713ddb0b07bd5839c",
"sha256": "9036343cc6af54f58c710505b130ca13ee9100f9cc88c513ab7383b49c423fa7"
},
"downloads": -1,
"filename": "pyember-0.0.16-cp310-cp310-macosx_11_0_arm64.whl",
"has_sig": false,
"md5_digest": "3f57a707b22d153713ddb0b07bd5839c",
"packagetype": "bdist_wheel",
"python_version": "cp310",
"requires_python": ">=3.9",
"size": 654751,
"upload_time": "2025-01-03T02:46:05",
"upload_time_iso_8601": "2025-01-03T02:46:05.695782Z",
"url": "https://files.pythonhosted.org/packages/e8/9a/d0c9abe9bb464b2703394d6fd0f4af74e58ccc5db26fbdcc15b1e69d4c4f/pyember-0.0.16-cp310-cp310-macosx_11_0_arm64.whl",
"yanked": false,
"yanked_reason": null
},
{
"comment_text": "",
"digests": {
"blake2b_256": "c61c88705f7ab6e27e582564fcdcec0d2284120b1362ec25d6cab0953a7e3def",
"md5": "94778e55a6d422f405daa07473fe2184",
"sha256": "bc2acc8dd6f3df88a1763e158dc8a3183ed9b9b9a0cea500f6de53f8c8ccecac"
},
"downloads": -1,
"filename": "pyember-0.0.16-cp310-cp310-win_amd64.whl",
"has_sig": false,
"md5_digest": "94778e55a6d422f405daa07473fe2184",
"packagetype": "bdist_wheel",
"python_version": "cp310",
"requires_python": ">=3.9",
"size": 234452,
"upload_time": "2025-01-03T02:46:06",
"upload_time_iso_8601": "2025-01-03T02:46:06.917601Z",
"url": "https://files.pythonhosted.org/packages/c6/1c/88705f7ab6e27e582564fcdcec0d2284120b1362ec25d6cab0953a7e3def/pyember-0.0.16-cp310-cp310-win_amd64.whl",
"yanked": false,
"yanked_reason": null
},
{
"comment_text": "",
"digests": {
"blake2b_256": "162977ac4ca61605dd49145fb2eb3260a4f49570fbafd83f1993ea353694a857",
"md5": "281b55f07380bcfe590c1a644dd652af",
"sha256": "370bede977c51a99babcd4b2c5c36282b3049020e77779aee7ed5c2dd7e98a6b"
},
"downloads": -1,
"filename": "pyember-0.0.16-cp311-cp311-macosx_11_0_arm64.whl",
"has_sig": false,
"md5_digest": "281b55f07380bcfe590c1a644dd652af",
"packagetype": "bdist_wheel",
"python_version": "cp311",
"requires_python": ">=3.9",
"size": 657117,
"upload_time": "2025-01-03T02:46:08",
"upload_time_iso_8601": "2025-01-03T02:46:08.019847Z",
"url": "https://files.pythonhosted.org/packages/16/29/77ac4ca61605dd49145fb2eb3260a4f49570fbafd83f1993ea353694a857/pyember-0.0.16-cp311-cp311-macosx_11_0_arm64.whl",
"yanked": false,
"yanked_reason": null
},
{
"comment_text": "",
"digests": {
"blake2b_256": "f5939c710dd8bd3c3dde18f820dc2aacb9c14083330ea73275f988c020643697",
"md5": "20ae2674604c5a021da707ffa0abbb09",
"sha256": "819e696a72fa57e69495dd6218a28c01cdbadefd3d6f097834bd83b118a8ae6d"
},
"downloads": -1,
"filename": "pyember-0.0.16-cp311-cp311-manylinux_2_27_x86_64.manylinux_2_28_x86_64.whl",
"has_sig": false,
"md5_digest": "20ae2674604c5a021da707ffa0abbb09",
"packagetype": "bdist_wheel",
"python_version": "cp311",
"requires_python": ">=3.9",
"size": 391154,
"upload_time": "2025-01-03T02:46:09",
"upload_time_iso_8601": "2025-01-03T02:46:09.388637Z",
"url": "https://files.pythonhosted.org/packages/f5/93/9c710dd8bd3c3dde18f820dc2aacb9c14083330ea73275f988c020643697/pyember-0.0.16-cp311-cp311-manylinux_2_27_x86_64.manylinux_2_28_x86_64.whl",
"yanked": false,
"yanked_reason": null
},
{
"comment_text": "",
"digests": {
"blake2b_256": "ea6be94b487c8bd99c7f30449721bbbbb29aa6ac34ed5966110c4f39ec2e106f",
"md5": "34a66e0d3bc70f479fb51ffb9c65eb3f",
"sha256": "02ddd4c5dc9574b3bb13905efca74ce4f89ba452e815729b8a3b5f9c6e3272ef"
},
"downloads": -1,
"filename": "pyember-0.0.16-cp311-cp311-win_amd64.whl",
"has_sig": false,
"md5_digest": "34a66e0d3bc70f479fb51ffb9c65eb3f",
"packagetype": "bdist_wheel",
"python_version": "cp311",
"requires_python": ">=3.9",
"size": 235838,
"upload_time": "2025-01-03T02:46:10",
"upload_time_iso_8601": "2025-01-03T02:46:10.510564Z",
"url": "https://files.pythonhosted.org/packages/ea/6b/e94b487c8bd99c7f30449721bbbbb29aa6ac34ed5966110c4f39ec2e106f/pyember-0.0.16-cp311-cp311-win_amd64.whl",
"yanked": false,
"yanked_reason": null
},
{
"comment_text": "",
"digests": {
"blake2b_256": "052010558f89b5b14be35b14b266a197ebca976fabcadee805f9bf73d2235cff",
"md5": "dc5829a3421fdb45391188b39e0781d2",
"sha256": "037864eaf7d9081287ab1b71fa08d08601bea55114b28811ad5cece19a297b27"
},
"downloads": -1,
"filename": "pyember-0.0.16-cp312-cp312-macosx_11_0_arm64.whl",
"has_sig": false,
"md5_digest": "dc5829a3421fdb45391188b39e0781d2",
"packagetype": "bdist_wheel",
"python_version": "cp312",
"requires_python": ">=3.9",
"size": 654324,
"upload_time": "2025-01-03T02:46:11",
"upload_time_iso_8601": "2025-01-03T02:46:11.688040Z",
"url": "https://files.pythonhosted.org/packages/05/20/10558f89b5b14be35b14b266a197ebca976fabcadee805f9bf73d2235cff/pyember-0.0.16-cp312-cp312-macosx_11_0_arm64.whl",
"yanked": false,
"yanked_reason": null
},
{
"comment_text": "",
"digests": {
"blake2b_256": "641cc2082e5497515684ff13e129d8323488574afd03855994a33170fcd64360",
"md5": "549872e37ad32c8d17e65b222a774a32",
"sha256": "b3154005cbfc05d4c928c62e7b6b1988a83d62fe35beba45a348fdcab916c843"
},
"downloads": -1,
"filename": "pyember-0.0.16-cp312-cp312-manylinux_2_27_x86_64.manylinux_2_28_x86_64.whl",
"has_sig": false,
"md5_digest": "549872e37ad32c8d17e65b222a774a32",
"packagetype": "bdist_wheel",
"python_version": "cp312",
"requires_python": ">=3.9",
"size": 390784,
"upload_time": "2025-01-03T02:46:12",
"upload_time_iso_8601": "2025-01-03T02:46:12.856322Z",
"url": "https://files.pythonhosted.org/packages/64/1c/c2082e5497515684ff13e129d8323488574afd03855994a33170fcd64360/pyember-0.0.16-cp312-cp312-manylinux_2_27_x86_64.manylinux_2_28_x86_64.whl",
"yanked": false,
"yanked_reason": null
},
{
"comment_text": "",
"digests": {
"blake2b_256": "795d25543debfc768ad180443a696435d2274944c97cc671462df2d0a12eaecf",
"md5": "9ebe8e62cedfae50dbf78a4fd898c1aa",
"sha256": "57abfcbe2ff473f84865d76c7104a99acddce30d7cc27b2e8d902d722cf81961"
},
"downloads": -1,
"filename": "pyember-0.0.16-cp312-cp312-win_amd64.whl",
"has_sig": false,
"md5_digest": "9ebe8e62cedfae50dbf78a4fd898c1aa",
"packagetype": "bdist_wheel",
"python_version": "cp312",
"requires_python": ">=3.9",
"size": 237128,
"upload_time": "2025-01-03T02:46:14",
"upload_time_iso_8601": "2025-01-03T02:46:14.031203Z",
"url": "https://files.pythonhosted.org/packages/79/5d/25543debfc768ad180443a696435d2274944c97cc671462df2d0a12eaecf/pyember-0.0.16-cp312-cp312-win_amd64.whl",
"yanked": false,
"yanked_reason": null
},
{
"comment_text": "",
"digests": {
"blake2b_256": "b019bc6e797cd5aad7036715d1d473b0fc95c00c116020cf0044e258621e5190",
"md5": "697297ff270f9cc36705dfa19c45260f",
"sha256": "8c552fb16596fc75374e01242334f1520c7de1a70811de36ac4d63df180191ba"
},
"downloads": -1,
"filename": "pyember-0.0.16-cp313-cp313-macosx_11_0_arm64.whl",
"has_sig": false,
"md5_digest": "697297ff270f9cc36705dfa19c45260f",
"packagetype": "bdist_wheel",
"python_version": "cp313",
"requires_python": ">=3.9",
"size": 654877,
"upload_time": "2025-01-03T02:46:15",
"upload_time_iso_8601": "2025-01-03T02:46:15.597311Z",
"url": "https://files.pythonhosted.org/packages/b0/19/bc6e797cd5aad7036715d1d473b0fc95c00c116020cf0044e258621e5190/pyember-0.0.16-cp313-cp313-macosx_11_0_arm64.whl",
"yanked": false,
"yanked_reason": null
},
{
"comment_text": "",
"digests": {
"blake2b_256": "75a785aaf2660d5f9acc746e8b320f3167104892948442806be8328cb8e7834b",
"md5": "b4c90471da85d6da90ba261464393e42",
"sha256": "d02973117f82dc64605dd857371564794e377f2b32b89cde74eb26b3d0202808"
},
"downloads": -1,
"filename": "pyember-0.0.16-cp313-cp313-win_amd64.whl",
"has_sig": false,
"md5_digest": "b4c90471da85d6da90ba261464393e42",
"packagetype": "bdist_wheel",
"python_version": "cp313",
"requires_python": ">=3.9",
"size": 237189,
"upload_time": "2025-01-03T02:46:18",
"upload_time_iso_8601": "2025-01-03T02:46:18.021413Z",
"url": "https://files.pythonhosted.org/packages/75/a7/85aaf2660d5f9acc746e8b320f3167104892948442806be8328cb8e7834b/pyember-0.0.16-cp313-cp313-win_amd64.whl",
"yanked": false,
"yanked_reason": null
},
{
"comment_text": "",
"digests": {
"blake2b_256": "42a915fbc0c94cd8f5090ddd31ee6c4e031dd4f57f5e0cfa092a691887674aab",
"md5": "2389530377f9a920ade83ba25c8e7a53",
"sha256": "f935035904fdd4dd96ef46751852cf796a93494262d04264867e2be9bcbbd1f2"
},
"downloads": -1,
"filename": "pyember-0.0.16-cp39-cp39-macosx_11_0_arm64.whl",
"has_sig": false,
"md5_digest": "2389530377f9a920ade83ba25c8e7a53",
"packagetype": "bdist_wheel",
"python_version": "cp39",
"requires_python": ">=3.9",
"size": 655082,
"upload_time": "2025-01-03T02:46:19",
"upload_time_iso_8601": "2025-01-03T02:46:19.152627Z",
"url": "https://files.pythonhosted.org/packages/42/a9/15fbc0c94cd8f5090ddd31ee6c4e031dd4f57f5e0cfa092a691887674aab/pyember-0.0.16-cp39-cp39-macosx_11_0_arm64.whl",
"yanked": false,
"yanked_reason": null
},
{
"comment_text": "",
"digests": {
"blake2b_256": "4d5458dc6aeb1cf990c419fe4760adf188ce90a50afff04978801c6d1482430e",
"md5": "fd1548c6aebbe2572a2334422a424616",
"sha256": "43469054a7319cf3ba160ffd9f6493697adc62f61bd8510c0e127db2c9dba071"
},
"downloads": -1,
"filename": "pyember-0.0.16-cp39-cp39-win_amd64.whl",
"has_sig": false,
"md5_digest": "fd1548c6aebbe2572a2334422a424616",
"packagetype": "bdist_wheel",
"python_version": "cp39",
"requires_python": ">=3.9",
"size": 228339,
"upload_time": "2025-01-03T02:46:20",
"upload_time_iso_8601": "2025-01-03T02:46:20.487876Z",
"url": "https://files.pythonhosted.org/packages/4d/54/58dc6aeb1cf990c419fe4760adf188ce90a50afff04978801c6d1482430e/pyember-0.0.16-cp39-cp39-win_amd64.whl",
"yanked": false,
"yanked_reason": null
}
],
"upload_time": "2025-01-03 02:46:05",
"github": true,
"gitlab": false,
"bitbucket": false,
"codeberg": false,
"github_user": "mbahng",
"github_project": "pyember",
"travis_ci": false,
"coveralls": false,
"github_actions": true,
"lcname": "pyember"
}
JSON
