phygnn binary classification example¶
Here's an example of how to make a binary classifier using a phygnn model without a p_fun.
In [1]:
import numpy as np
import pandas as pd
import matplotlib.pyplot as plt
import tensorflow as tf
from rex import init_logger
from phygnn import PhysicsGuidedNeuralNetwork, p_fun_dummy
In [2]:
init_logger('phygnn', log_level='INFO', log_file=None)
Out[2]:
<Logger phygnn (INFO)>
In [3]:
x1 = np.arange(500) - 250
x2 = np.arange(500) - 250
x1, x2 = np.meshgrid(x1, x2)
x1 = x1.flatten()
x2 = x2.flatten()
x3 = x1 * x2
features = pd.DataFrame({'x1': x1, 'x2': x2})
features.head()
Out[3]:
| x1 | x2 | |
|---|---|---|
| 0 | -250 | -250 |
| 1 | -249 | -250 |
| 2 | -248 | -250 |
| 3 | -247 | -250 |
| 4 | -246 | -250 |
In [4]:
y = ((x1 * x2) > 0).astype(bool).astype(float)
labels = pd.DataFrame({'y': y})
labels.head()
Out[4]:
| y | |
|---|---|
| 0 | 1.0 |
| 1 | 1.0 |
| 2 | 1.0 |
| 3 | 1.0 |
| 4 | 1.0 |
In [5]:
hidden_layers = [{'units': 16},
{'activation': 'relu'},
{'units': 16},
{'activation': 'relu'},
]
output_layer = [{'units': 1},
{'activation': 'sigmoid'},
]
In [6]:
PhysicsGuidedNeuralNetwork.seed(0)
model = PhysicsGuidedNeuralNetwork(p_fun=p_fun_dummy,
hidden_layers=hidden_layers,
output_layer=output_layer,
loss_weights=(1.0, 0.0),
metric='binary_crossentropy',
learning_rate=0.05,
n_features=2, n_labels=1)
model.fit(features, labels, features, n_batch=1, n_epoch=50)
INFO - 2021-02-02 16:35:44,812 [phygnn.py:773] : Epoch 0 train loss: 1.26e+01 val loss: 8.18e+00 for "phygnn" INFO - 2021-02-02 16:35:45,011 [phygnn.py:773] : Epoch 1 train loss: 8.26e+00 val loss: 1.73e+00 for "phygnn" INFO - 2021-02-02 16:35:45,189 [phygnn.py:773] : Epoch 2 train loss: 1.75e+00 val loss: 3.29e+00 for "phygnn" INFO - 2021-02-02 16:35:45,334 [phygnn.py:773] : Epoch 3 train loss: 3.31e+00 val loss: 3.03e+00 for "phygnn" INFO - 2021-02-02 16:35:45,505 [phygnn.py:773] : Epoch 4 train loss: 3.06e+00 val loss: 1.59e+00 for "phygnn" INFO - 2021-02-02 16:35:45,651 [phygnn.py:773] : Epoch 5 train loss: 1.63e+00 val loss: 6.06e-01 for "phygnn" INFO - 2021-02-02 16:35:45,817 [phygnn.py:773] : Epoch 6 train loss: 6.22e-01 val loss: 1.46e-01 for "phygnn" INFO - 2021-02-02 16:35:45,995 [phygnn.py:773] : Epoch 7 train loss: 1.44e-01 val loss: 1.58e-01 for "phygnn" INFO - 2021-02-02 16:35:46,180 [phygnn.py:773] : Epoch 8 train loss: 1.55e-01 val loss: 3.75e-01 for "phygnn" INFO - 2021-02-02 16:35:46,367 [phygnn.py:773] : Epoch 9 train loss: 3.59e-01 val loss: 4.78e-01 for "phygnn" INFO - 2021-02-02 16:35:46,571 [phygnn.py:773] : Epoch 10 train loss: 4.50e-01 val loss: 4.42e-01 for "phygnn" INFO - 2021-02-02 16:35:46,731 [phygnn.py:773] : Epoch 11 train loss: 4.09e-01 val loss: 4.50e-01 for "phygnn" INFO - 2021-02-02 16:35:46,882 [phygnn.py:773] : Epoch 12 train loss: 4.17e-01 val loss: 3.86e-01 for "phygnn" INFO - 2021-02-02 16:35:47,051 [phygnn.py:773] : Epoch 13 train loss: 3.60e-01 val loss: 3.39e-01 for "phygnn" INFO - 2021-02-02 16:35:47,196 [phygnn.py:773] : Epoch 14 train loss: 3.25e-01 val loss: 2.54e-01 for "phygnn" INFO - 2021-02-02 16:35:47,338 [phygnn.py:773] : Epoch 15 train loss: 2.46e-01 val loss: 1.08e-01 for "phygnn" INFO - 2021-02-02 16:35:47,475 [phygnn.py:773] : Epoch 16 train loss: 1.03e-01 val loss: 9.77e-02 for "phygnn" INFO - 2021-02-02 16:35:47,620 [phygnn.py:773] : Epoch 17 train loss: 9.36e-02 val loss: 6.76e-02 for "phygnn" INFO - 2021-02-02 16:35:47,776 [phygnn.py:773] : Epoch 18 train loss: 6.63e-02 val loss: 1.33e-01 for "phygnn" INFO - 2021-02-02 16:35:47,916 [phygnn.py:773] : Epoch 19 train loss: 1.33e-01 val loss: 6.92e-02 for "phygnn" INFO - 2021-02-02 16:35:48,058 [phygnn.py:773] : Epoch 20 train loss: 6.99e-02 val loss: 7.40e-02 for "phygnn" INFO - 2021-02-02 16:35:48,217 [phygnn.py:773] : Epoch 21 train loss: 7.37e-02 val loss: 7.44e-02 for "phygnn" INFO - 2021-02-02 16:35:48,373 [phygnn.py:773] : Epoch 22 train loss: 7.48e-02 val loss: 6.11e-02 for "phygnn" INFO - 2021-02-02 16:35:48,518 [phygnn.py:773] : Epoch 23 train loss: 6.13e-02 val loss: 5.84e-02 for "phygnn" INFO - 2021-02-02 16:35:48,654 [phygnn.py:773] : Epoch 24 train loss: 5.72e-02 val loss: 4.71e-02 for "phygnn" INFO - 2021-02-02 16:35:48,794 [phygnn.py:773] : Epoch 25 train loss: 4.42e-02 val loss: 8.04e-02 for "phygnn" INFO - 2021-02-02 16:35:48,929 [phygnn.py:773] : Epoch 26 train loss: 7.55e-02 val loss: 5.56e-02 for "phygnn" INFO - 2021-02-02 16:35:49,067 [phygnn.py:773] : Epoch 27 train loss: 5.27e-02 val loss: 5.29e-02 for "phygnn" INFO - 2021-02-02 16:35:49,209 [phygnn.py:773] : Epoch 28 train loss: 5.14e-02 val loss: 2.71e-02 for "phygnn" INFO - 2021-02-02 16:35:49,348 [phygnn.py:773] : Epoch 29 train loss: 2.64e-02 val loss: 3.21e-02 for "phygnn" INFO - 2021-02-02 16:35:49,482 [phygnn.py:773] : Epoch 30 train loss: 3.15e-02 val loss: 3.60e-02 for "phygnn" INFO - 2021-02-02 16:35:49,614 [phygnn.py:773] : Epoch 31 train loss: 3.62e-02 val loss: 3.07e-02 for "phygnn" INFO - 2021-02-02 16:35:49,747 [phygnn.py:773] : Epoch 32 train loss: 3.20e-02 val loss: 4.04e-02 for "phygnn" INFO - 2021-02-02 16:35:49,889 [phygnn.py:773] : Epoch 33 train loss: 4.18e-02 val loss: 3.16e-02 for "phygnn" INFO - 2021-02-02 16:35:50,045 [phygnn.py:773] : Epoch 34 train loss: 3.29e-02 val loss: 3.54e-02 for "phygnn" INFO - 2021-02-02 16:35:50,240 [phygnn.py:773] : Epoch 35 train loss: 3.54e-02 val loss: 2.00e-02 for "phygnn" INFO - 2021-02-02 16:35:50,413 [phygnn.py:773] : Epoch 36 train loss: 2.03e-02 val loss: 2.21e-02 for "phygnn" INFO - 2021-02-02 16:35:50,550 [phygnn.py:773] : Epoch 37 train loss: 2.22e-02 val loss: 2.38e-02 for "phygnn" INFO - 2021-02-02 16:35:50,691 [phygnn.py:773] : Epoch 38 train loss: 2.34e-02 val loss: 2.39e-02 for "phygnn" INFO - 2021-02-02 16:35:50,839 [phygnn.py:773] : Epoch 39 train loss: 2.30e-02 val loss: 2.59e-02 for "phygnn" INFO - 2021-02-02 16:35:50,972 [phygnn.py:773] : Epoch 40 train loss: 2.44e-02 val loss: 2.26e-02 for "phygnn" INFO - 2021-02-02 16:35:51,121 [phygnn.py:773] : Epoch 41 train loss: 2.17e-02 val loss: 2.71e-02 for "phygnn" INFO - 2021-02-02 16:35:51,262 [phygnn.py:773] : Epoch 42 train loss: 2.63e-02 val loss: 1.94e-02 for "phygnn" INFO - 2021-02-02 16:35:51,399 [phygnn.py:773] : Epoch 43 train loss: 1.89e-02 val loss: 1.86e-02 for "phygnn" INFO - 2021-02-02 16:35:51,535 [phygnn.py:773] : Epoch 44 train loss: 1.77e-02 val loss: 1.62e-02 for "phygnn" INFO - 2021-02-02 16:35:51,666 [phygnn.py:773] : Epoch 45 train loss: 1.59e-02 val loss: 1.83e-02 for "phygnn" INFO - 2021-02-02 16:35:51,806 [phygnn.py:773] : Epoch 46 train loss: 1.86e-02 val loss: 1.82e-02 for "phygnn" INFO - 2021-02-02 16:35:51,937 [phygnn.py:773] : Epoch 47 train loss: 1.87e-02 val loss: 1.65e-02 for "phygnn" INFO - 2021-02-02 16:35:52,071 [phygnn.py:773] : Epoch 48 train loss: 1.70e-02 val loss: 1.85e-02 for "phygnn" INFO - 2021-02-02 16:35:52,209 [phygnn.py:773] : Epoch 49 train loss: 1.87e-02 val loss: 1.64e-02 for "phygnn"
In [7]:
model.history[['training_loss', 'validation_loss']].plot()
plt.ylabel('Loss')
plt.show()
plt.close()
y_pred = model.predict(features)
accuracy = 100 * (np.round(y_pred) == labels.values).sum() / len(labels)
print('Accuracy: {:.2f}%'.format(accuracy))
Accuracy: 99.49%
In [ ]:
In [ ]: