Polynomial chaos expansion (PCE)#

A PCERegressor is a PCE model based on OpenTURNS.

from __future__ import annotations

from matplotlib import pyplot as plt
from numpy import array

from gemseo import configure_logger
from gemseo import create_discipline
from gemseo import create_parameter_space
from gemseo import sample_disciplines
from gemseo.mlearning import create_regression_model

configure_logger()
<RootLogger root (INFO)>

Problem#

In this example, we represent the function \(f(x)=(6x-2)^2\sin(12x-4)\) [FSK08] by the AnalyticDiscipline

discipline = create_discipline(
    "AnalyticDiscipline",
    name="f",
    expressions={"y": "(6*x-2)**2*sin(12*x-4)"},
)

and seek to approximate it over the input space

input_space = create_parameter_space()
input_space.add_random_variable("x", "OTUniformDistribution")

To do this, we create a training dataset with 6 equispaced points:

training_dataset = sample_disciplines(
    [discipline], input_space, "y", algo_name="PYDOE_FULLFACT", n_samples=10
)
WARNING - 00:09:41: No coupling in MDA, switching chain_linearize to True.
   INFO - 00:09:41: *** Start Sampling execution ***
   INFO - 00:09:41: Sampling
   INFO - 00:09:41:    Disciplines: f
   INFO - 00:09:41:    MDO formulation: MDF
   INFO - 00:09:41: Running the algorithm PYDOE_FULLFACT:
   INFO - 00:09:41:     10%|█         | 1/10 [00:00<00:00, 669.91 it/sec]
   INFO - 00:09:41:     20%|██        | 2/10 [00:00<00:00, 1079.34 it/sec]
   INFO - 00:09:41:     30%|███       | 3/10 [00:00<00:00, 1408.27 it/sec]
   INFO - 00:09:41:     40%|████      | 4/10 [00:00<00:00, 1677.89 it/sec]
   INFO - 00:09:41:     50%|█████     | 5/10 [00:00<00:00, 1901.83 it/sec]
   INFO - 00:09:41:     60%|██████    | 6/10 [00:00<00:00, 2076.73 it/sec]
   INFO - 00:09:41:     70%|███████   | 7/10 [00:00<00:00, 2232.71 it/sec]
   INFO - 00:09:41:     80%|████████  | 8/10 [00:00<00:00, 2375.54 it/sec]
   INFO - 00:09:41:     90%|█████████ | 9/10 [00:00<00:00, 2504.23 it/sec]
   INFO - 00:09:41:    100%|██████████| 10/10 [00:00<00:00, 2605.16 it/sec]
   INFO - 00:09:41: *** End Sampling execution (time: 0:00:00.005014) ***

Basics#

Training#

Then, we train an PCE regression model from these samples:

model = create_regression_model("PCERegressor", training_dataset)
model.learn()
WARNING - 00:09:41: Remove input data transformation because PCERegressor does not support transformers.

Prediction#

Once it is built, we can predict the output value of \(f\) at a new input point:

input_value = {"x": array([0.65])}
output_value = model.predict(input_value)
output_value
{'y': array([-0.81106394])}

as well as its Jacobian value:

jacobian_value = model.predict_jacobian(input_value)
jacobian_value
{'y': {'x': array([[18.2279622]])}}

Plotting#

Of course, you can see that the quadratic model is no good at all here:

test_dataset = sample_disciplines(
    [discipline], input_space, "y", algo_name="PYDOE_FULLFACT", n_samples=100
)
input_data = test_dataset.get_view(variable_names=model.input_names).to_numpy()
reference_output_data = test_dataset.get_view(variable_names="y").to_numpy().ravel()
predicted_output_data = model.predict(input_data).ravel()
plt.plot(input_data.ravel(), reference_output_data, label="Reference")
plt.plot(input_data.ravel(), predicted_output_data, label="Regression - Basics")
plt.grid()
plt.legend()
plt.show()
plot pce regression
WARNING - 00:09:41: No coupling in MDA, switching chain_linearize to True.
   INFO - 00:09:41: *** Start Sampling execution ***
   INFO - 00:09:41: Sampling
   INFO - 00:09:41:    Disciplines: f
   INFO - 00:09:41:    MDO formulation: MDF
   INFO - 00:09:41: Running the algorithm PYDOE_FULLFACT:
   INFO - 00:09:41:      1%|          | 1/100 [00:00<00:00, 3457.79 it/sec]
   INFO - 00:09:41:      2%|▏         | 2/100 [00:00<00:00, 3298.71 it/sec]
   INFO - 00:09:41:      3%|▎         | 3/100 [00:00<00:00, 3456.84 it/sec]
   INFO - 00:09:41:      4%|▍         | 4/100 [00:00<00:00, 3590.25 it/sec]
   INFO - 00:09:41:      5%|▌         | 5/100 [00:00<00:00, 3687.62 it/sec]
   INFO - 00:09:41:      6%|▌         | 6/100 [00:00<00:00, 3705.22 it/sec]
   INFO - 00:09:41:      7%|▋         | 7/100 [00:00<00:00, 3770.40 it/sec]
   INFO - 00:09:41:      8%|▊         | 8/100 [00:00<00:00, 3824.74 it/sec]
   INFO - 00:09:41:      9%|▉         | 9/100 [00:00<00:00, 3871.27 it/sec]
   INFO - 00:09:41:     10%|█         | 10/100 [00:00<00:00, 3871.43 it/sec]
   INFO - 00:09:41:     11%|█         | 11/100 [00:00<00:00, 3903.99 it/sec]
   INFO - 00:09:41:     12%|█▏        | 12/100 [00:00<00:00, 3926.03 it/sec]
   INFO - 00:09:41:     13%|█▎        | 13/100 [00:00<00:00, 3954.02 it/sec]
   INFO - 00:09:41:     14%|█▍        | 14/100 [00:00<00:00, 3956.89 it/sec]
   INFO - 00:09:41:     15%|█▌        | 15/100 [00:00<00:00, 3979.42 it/sec]
   INFO - 00:09:41:     16%|█▌        | 16/100 [00:00<00:00, 4002.44 it/sec]
   INFO - 00:09:41:     17%|█▋        | 17/100 [00:00<00:00, 4025.47 it/sec]
   INFO - 00:09:41:     18%|█▊        | 18/100 [00:00<00:00, 4025.89 it/sec]
   INFO - 00:09:41:     19%|█▉        | 19/100 [00:00<00:00, 4032.17 it/sec]
   INFO - 00:09:41:     20%|██        | 20/100 [00:00<00:00, 4046.41 it/sec]
   INFO - 00:09:41:     21%|██        | 21/100 [00:00<00:00, 4060.50 it/sec]
   INFO - 00:09:41:     22%|██▏       | 22/100 [00:00<00:00, 4075.92 it/sec]
   INFO - 00:09:41:     23%|██▎       | 23/100 [00:00<00:00, 4072.31 it/sec]
   INFO - 00:09:41:     24%|██▍       | 24/100 [00:00<00:00, 4086.85 it/sec]
   INFO - 00:09:41:     25%|██▌       | 25/100 [00:00<00:00, 4098.56 it/sec]
   INFO - 00:09:41:     26%|██▌       | 26/100 [00:00<00:00, 4109.58 it/sec]
   INFO - 00:09:41:     27%|██▋       | 27/100 [00:00<00:00, 4101.49 it/sec]
   INFO - 00:09:41:     28%|██▊       | 28/100 [00:00<00:00, 4107.89 it/sec]
   INFO - 00:09:41:     29%|██▉       | 29/100 [00:00<00:00, 4112.06 it/sec]
   INFO - 00:09:41:     30%|███       | 30/100 [00:00<00:00, 4120.95 it/sec]
   INFO - 00:09:41:     31%|███       | 31/100 [00:00<00:00, 4104.02 it/sec]
   INFO - 00:09:41:     32%|███▏      | 32/100 [00:00<00:00, 4110.30 it/sec]
   INFO - 00:09:41:     33%|███▎      | 33/100 [00:00<00:00, 4120.02 it/sec]
   INFO - 00:09:41:     34%|███▍      | 34/100 [00:00<00:00, 4129.21 it/sec]
   INFO - 00:09:41:     35%|███▌      | 35/100 [00:00<00:00, 4125.93 it/sec]
   INFO - 00:09:41:     36%|███▌      | 36/100 [00:00<00:00, 4131.75 it/sec]
   INFO - 00:09:41:     37%|███▋      | 37/100 [00:00<00:00, 4140.04 it/sec]
   INFO - 00:09:41:     38%|███▊      | 38/100 [00:00<00:00, 4148.99 it/sec]
   INFO - 00:09:41:     39%|███▉      | 39/100 [00:00<00:00, 4158.16 it/sec]
   INFO - 00:09:41:     40%|████      | 40/100 [00:00<00:00, 4154.42 it/sec]
   INFO - 00:09:41:     41%|████      | 41/100 [00:00<00:00, 4161.72 it/sec]
   INFO - 00:09:41:     42%|████▏     | 42/100 [00:00<00:00, 4169.09 it/sec]
   INFO - 00:09:41:     43%|████▎     | 43/100 [00:00<00:00, 4175.08 it/sec]
   INFO - 00:09:41:     44%|████▍     | 44/100 [00:00<00:00, 4172.40 it/sec]
   INFO - 00:09:41:     45%|████▌     | 45/100 [00:00<00:00, 4176.30 it/sec]
   INFO - 00:09:41:     46%|████▌     | 46/100 [00:00<00:00, 4178.86 it/sec]
   INFO - 00:09:41:     47%|████▋     | 47/100 [00:00<00:00, 4183.27 it/sec]
   INFO - 00:09:41:     48%|████▊     | 48/100 [00:00<00:00, 4182.11 it/sec]
   INFO - 00:09:41:     49%|████▉     | 49/100 [00:00<00:00, 4184.14 it/sec]
   INFO - 00:09:41:     50%|█████     | 50/100 [00:00<00:00, 4190.03 it/sec]
   INFO - 00:09:41:     51%|█████     | 51/100 [00:00<00:00, 4195.46 it/sec]
   INFO - 00:09:41:     52%|█████▏    | 52/100 [00:00<00:00, 4200.28 it/sec]
   INFO - 00:09:41:     53%|█████▎    | 53/100 [00:00<00:00, 4195.97 it/sec]
   INFO - 00:09:41:     54%|█████▍    | 54/100 [00:00<00:00, 4200.06 it/sec]
   INFO - 00:09:41:     55%|█████▌    | 55/100 [00:00<00:00, 4204.78 it/sec]
   INFO - 00:09:41:     56%|█████▌    | 56/100 [00:00<00:00, 4210.24 it/sec]
   INFO - 00:09:41:     57%|█████▋    | 57/100 [00:00<00:00, 4208.33 it/sec]
   INFO - 00:09:41:     58%|█████▊    | 58/100 [00:00<00:00, 4211.80 it/sec]
   INFO - 00:09:41:     59%|█████▉    | 59/100 [00:00<00:00, 4215.02 it/sec]
   INFO - 00:09:41:     60%|██████    | 60/100 [00:00<00:00, 4219.27 it/sec]
   INFO - 00:09:41:     61%|██████    | 61/100 [00:00<00:00, 4217.26 it/sec]
   INFO - 00:09:41:     62%|██████▏   | 62/100 [00:00<00:00, 4190.86 it/sec]
   INFO - 00:09:41:     63%|██████▎   | 63/100 [00:00<00:00, 4189.85 it/sec]
   INFO - 00:09:41:     64%|██████▍   | 64/100 [00:00<00:00, 4192.27 it/sec]
   INFO - 00:09:41:     65%|██████▌   | 65/100 [00:00<00:00, 4190.05 it/sec]
   INFO - 00:09:41:     66%|██████▌   | 66/100 [00:00<00:00, 4192.65 it/sec]
   INFO - 00:09:41:     67%|██████▋   | 67/100 [00:00<00:00, 4195.43 it/sec]
   INFO - 00:09:41:     68%|██████▊   | 68/100 [00:00<00:00, 4199.37 it/sec]
   INFO - 00:09:41:     69%|██████▉   | 69/100 [00:00<00:00, 4199.23 it/sec]
   INFO - 00:09:41:     70%|███████   | 70/100 [00:00<00:00, 4200.85 it/sec]
   INFO - 00:09:41:     71%|███████   | 71/100 [00:00<00:00, 4204.25 it/sec]
   INFO - 00:09:41:     72%|███████▏  | 72/100 [00:00<00:00, 4207.92 it/sec]
   INFO - 00:09:41:     73%|███████▎  | 73/100 [00:00<00:00, 4210.40 it/sec]
   INFO - 00:09:41:     74%|███████▍  | 74/100 [00:00<00:00, 4207.04 it/sec]
   INFO - 00:09:41:     75%|███████▌  | 75/100 [00:00<00:00, 4209.46 it/sec]
   INFO - 00:09:41:     76%|███████▌  | 76/100 [00:00<00:00, 4212.48 it/sec]
   INFO - 00:09:41:     77%|███████▋  | 77/100 [00:00<00:00, 4214.94 it/sec]
   INFO - 00:09:41:     78%|███████▊  | 78/100 [00:00<00:00, 4213.43 it/sec]
   INFO - 00:09:41:     79%|███████▉  | 79/100 [00:00<00:00, 4213.13 it/sec]
   INFO - 00:09:41:     80%|████████  | 80/100 [00:00<00:00, 4213.69 it/sec]
   INFO - 00:09:41:     81%|████████  | 81/100 [00:00<00:00, 4214.75 it/sec]
   INFO - 00:09:41:     82%|████████▏ | 82/100 [00:00<00:00, 4212.95 it/sec]
   INFO - 00:09:41:     83%|████████▎ | 83/100 [00:00<00:00, 4214.87 it/sec]
   INFO - 00:09:41:     84%|████████▍ | 84/100 [00:00<00:00, 4217.35 it/sec]
   INFO - 00:09:41:     85%|████████▌ | 85/100 [00:00<00:00, 4219.77 it/sec]
   INFO - 00:09:41:     86%|████████▌ | 86/100 [00:00<00:00, 4221.55 it/sec]
   INFO - 00:09:41:     87%|████████▋ | 87/100 [00:00<00:00, 4218.40 it/sec]
   INFO - 00:09:41:     88%|████████▊ | 88/100 [00:00<00:00, 4220.01 it/sec]
   INFO - 00:09:41:     89%|████████▉ | 89/100 [00:00<00:00, 4222.29 it/sec]
   INFO - 00:09:41:     90%|█████████ | 90/100 [00:00<00:00, 4225.05 it/sec]
   INFO - 00:09:41:     91%|█████████ | 91/100 [00:00<00:00, 4224.15 it/sec]
   INFO - 00:09:41:     92%|█████████▏| 92/100 [00:00<00:00, 4226.00 it/sec]
   INFO - 00:09:41:     93%|█████████▎| 93/100 [00:00<00:00, 4227.90 it/sec]
   INFO - 00:09:41:     94%|█████████▍| 94/100 [00:00<00:00, 4229.17 it/sec]
   INFO - 00:09:41:     95%|█████████▌| 95/100 [00:00<00:00, 4227.77 it/sec]
   INFO - 00:09:41:     96%|█████████▌| 96/100 [00:00<00:00, 4228.93 it/sec]
   INFO - 00:09:41:     97%|█████████▋| 97/100 [00:00<00:00, 4231.38 it/sec]
   INFO - 00:09:41:     98%|█████████▊| 98/100 [00:00<00:00, 4232.00 it/sec]
   INFO - 00:09:41:     99%|█████████▉| 99/100 [00:00<00:00, 4234.90 it/sec]
   INFO - 00:09:41:    100%|██████████| 100/100 [00:00<00:00, 4233.16 it/sec]
   INFO - 00:09:41: *** End Sampling execution (time: 0:00:00.025480) ***

Settings#

The PCERegressor has many options defined in the PCERegressor_Settings Pydantic model.

Degree#

model = create_regression_model("PCERegressor", training_dataset, degree=3)
model.learn()
WARNING - 00:09:41: Remove input data transformation because PCERegressor does not support transformers.

and see that this model seems to be better:

predicted_output_data_ = model.predict(input_data).ravel()
plt.plot(input_data.ravel(), reference_output_data, label="Reference")
plt.plot(input_data.ravel(), predicted_output_data, label="Regression - Basics")
plt.plot(input_data.ravel(), predicted_output_data_, label="Regression - Degree(3)")
plt.grid()
plt.legend()
plt.show()
plot pce regression

Total running time of the script: (0 minutes 0.192 seconds)

Gallery generated by Sphinx-Gallery