Artifice 0.0.3

A homemade machine learning platform modeled after TensorFlow

Artifice

Artifice is a homemade machine learning platform modeled after TensorFlow, co-developed and co-written by William Bidle and Ilana Zane.

Installation

To get started with Artifice, copy the following command and paste it into your command line:

pip install Artifice

To test the installation, run the following code into your Python editor of choice:

from Artifice import Artifice

layer_sequence = [1,'ReLU', 2, 'sigmoid', 3]
loss_function = 'MSLE'

nn = Artifice.NN(layer_sequence, loss_function)

print('activation func library:\n', nn.activation_funcs_library, '\n')
print('loss func library:\n', nn.loss_funcs_library, '\n')
print('current weights:\n', nn.weights, '\n')
print('current activation functions:\n', nn.activation_funcs, '\n')
print('current loss function:\n', nn.loss_func_label, ':', nn.loss_func, '\n')
print('traing error:\n', nn.training_err, '\n')

If there are no errors, then you have successfully installed Artifice! The full list of functions, their usage, as well as some examples can be found within the Artifice.py file.

List of available activation functions

For a given value, $x$, different activation functions are definined by the following.

• "sigmoid" :

$$\frac{1}{1 + e^{-x}}$$

• 'tanh' :

$$tanh(x)$$

• 'ReLU' :

$$f(x) = \begin{cases} x & \text{if } x \geq 0,\ 0 & \text{if } x < 0. \end{cases}$$

List of avaliable loss functions

For a given network output vector, $\vec{y}^{out}$, and true value vector, $\vec{y}^{true}$, with $N$ components each, different loss functions are definined by the following.

• Mean Squared Error ("MSE") :

$$\sum_{i}^N(y_i^{out} - y_i^{true})^2$$

• Mean Absolute Error ("MAE") :

$$\sum_{i}^N|y_i^{out} - y_i^{true}|$$

• "MAPE" : $$100 * \sum_{i}^N|\frac{y_i^{out} - y_i^{true}}{y_i^{out} + y_i^{true}}|$$

• Mean Squared Logarithmic Error ("MSLE") :

$$\sum_{i}^N(log(y_i^{out} + 1) - log(y_i^{true} + 1))^2$$

• Binary Cross-Entropy ("BCE") :

$$\sum_{i}^N(y_i^{true}*log(y_i^{out}) + (1 - y_i^{true})*log(1 - y_i^{out}))$$

• "Poisson" :

$$\sum_{i}^N(y_i^{out} - y_i^{true} * log(y_i^{out}))$$

Examples

Detailed examples on how to use Artifice can be found in the Artifice_Tutorial.ipynb Jupyter Notebook.