nueramic-mathml 0.75

Math algorithms in ML on torch

---

Nueramic MathML

Nueramic-mathml is a library for visualizing and logging the steps of optimization algorithms in machine learning. The project uses torch for calculations and plotly for visualization.

pip install nueramic-mathml

Quick tour

Optimization

You can minimize the functions and see a detailed description of each step. After minimizing, you have a history with complete logs. Also available multidimensional optimisation.

def f(x): return x ** 3 - x ** 2 - x  # Minimum at x = 1
bounds = (0, 3)
one_optimize.golden_section_search(f, bounds, epsilon=0.01, verbose=True)[0]

Iteration: 0        |        point = 1.500  |        f(point) = -0.375
Iteration: 1        |        point = 0.927  |        f(point) = -0.990
Iteration: 2        |        point = 1.281  |        f(point) = -0.820
Iteration: 3        |        point = 1.062  |        f(point) = -0.992
Iteration: 4        |        point = 0.927  |        f(point) = -0.990
Iteration: 5        |        point = 1.011  |        f(point) = -1.000
Iteration: 6        |        point = 0.959  |        f(point) = -0.997
Iteration: 7        |        point = 0.991  |        f(point) = -1.000
Iteration: 8        |        point = 1.011  |        f(point) = -1.000
Iteration: 9        |        point = 0.998  |        f(point) = -1.000
Iteration: 10       |        point = 1.006  |        f(point) = -1.000
Searching finished. Successfully. code 0
1.0059846881033916

Models

You can use our models for classification and regression

from nueramic_mathml.ml import LogisticRegressionRBF
from sklearn.datasets import make_moons

x, y = make_moons(10_000, noise=.1, random_state=84)
x, y = torch.tensor(x), torch.tensor(y)
logistic_model_rbf = LogisticRegressionRBF(x[:1000]).fit(x, y, show_epoch=10)

Epoch:     1 | CrossEntropyLoss:  0.71496
Epoch:    12 | CrossEntropyLoss:  0.35328
Epoch:    23 | CrossEntropyLoss:  0.27769
Epoch:    34 | CrossEntropyLoss:  0.22395
Epoch:    45 | CrossEntropyLoss:  0.19266
Epoch:    56 | CrossEntropyLoss:  0.16695
Epoch:    67 | CrossEntropyLoss:  0.14686
Epoch:    78 | CrossEntropyLoss:  0.13051
Epoch:    89 | CrossEntropyLoss:  0.11724
Epoch:   100 | CrossEntropyLoss:  0.10629

logistic_model_rbf.metrics_tab(x, y)

{'auc_roc': 0.9974513817072977,
 'f1': 0.9700730618209839,
 'precision': 0.9709476828575134,
 'recall': 0.9692000150680542}

Visualizations

You can create beautiful animations of optimization algorithms and regression/classification models.

gen_classification_plot(x, y, model, threshold=0.5, epsilon=0.001)