Note

Go to the end to download the full example code.

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

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
)

INFO - 16:22:21: *** Start Sampling execution ***
INFO - 16:22:21: Sampling
INFO - 16:22:21:    Disciplines: f
INFO - 16:22:21:    MDO formulation: MDF
INFO - 16:22:21: Running the algorithm PYDOE_FULLFACT:
INFO - 16:22:21:     10%|█         | 1/10 [00:00<00:00, 711.99 it/sec]
INFO - 16:22:21:     20%|██        | 2/10 [00:00<00:00, 1161.21 it/sec]
INFO - 16:22:21:     30%|███       | 3/10 [00:00<00:00, 1494.76 it/sec]
INFO - 16:22:21:     40%|████      | 4/10 [00:00<00:00, 1771.99 it/sec]
INFO - 16:22:21:     50%|█████     | 5/10 [00:00<00:00, 1994.44 it/sec]
INFO - 16:22:21:     60%|██████    | 6/10 [00:00<00:00, 2187.57 it/sec]
INFO - 16:22:21:     70%|███████   | 7/10 [00:00<00:00, 2337.78 it/sec]
INFO - 16:22:21:     80%|████████  | 8/10 [00:00<00:00, 2480.00 it/sec]
INFO - 16:22:21:     90%|█████████ | 9/10 [00:00<00:00, 2608.76 it/sec]
INFO - 16:22:21:    100%|██████████| 10/10 [00:00<00:00, 2630.65 it/sec]
INFO - 16:22:21: *** End Sampling execution ***

Basics#

Training#

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

model = create_regression_model("PCERegressor", training_dataset)
model.learn()

WARNING - 16:22:21: 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
INFO - 16:22:21: *** Start Sampling execution ***
INFO - 16:22:21: Sampling
INFO - 16:22:21:    Disciplines: f
INFO - 16:22:21:    MDO formulation: MDF
INFO - 16:22:21: Running the algorithm PYDOE_FULLFACT:
INFO - 16:22:21:      1%|          | 1/100 [00:00<00:00, 4198.50 it/sec]
INFO - 16:22:21:      2%|▏         | 2/100 [00:00<00:00, 3479.31 it/sec]
INFO - 16:22:21:      3%|▎         | 3/100 [00:00<00:00, 3416.48 it/sec]
INFO - 16:22:21:      4%|▍         | 4/100 [00:00<00:00, 3519.45 it/sec]
INFO - 16:22:21:      5%|▌         | 5/100 [00:00<00:00, 3580.59 it/sec]
INFO - 16:22:21:      6%|▌         | 6/100 [00:00<00:00, 3611.11 it/sec]
INFO - 16:22:21:      7%|▋         | 7/100 [00:00<00:00, 3645.86 it/sec]
INFO - 16:22:21:      8%|▊         | 8/100 [00:00<00:00, 3690.14 it/sec]
INFO - 16:22:21:      9%|▉         | 9/100 [00:00<00:00, 3763.21 it/sec]
INFO - 16:22:21:     10%|█         | 10/100 [00:00<00:00, 3830.76 it/sec]
INFO - 16:22:21:     11%|█         | 11/100 [00:00<00:00, 3847.34 it/sec]
INFO - 16:22:21:     12%|█▏        | 12/100 [00:00<00:00, 3892.32 it/sec]
INFO - 16:22:21:     13%|█▎        | 13/100 [00:00<00:00, 3919.63 it/sec]
INFO - 16:22:21:     14%|█▍        | 14/100 [00:00<00:00, 3957.42 it/sec]
INFO - 16:22:21:     15%|█▌        | 15/100 [00:00<00:00, 3973.38 it/sec]
INFO - 16:22:21:     16%|█▌        | 16/100 [00:00<00:00, 4001.01 it/sec]
INFO - 16:22:21:     17%|█▋        | 17/100 [00:00<00:00, 4030.25 it/sec]
INFO - 16:22:21:     18%|█▊        | 18/100 [00:00<00:00, 4059.00 it/sec]
INFO - 16:22:21:     19%|█▉        | 19/100 [00:00<00:00, 4068.81 it/sec]
INFO - 16:22:21:     20%|██        | 20/100 [00:00<00:00, 4084.83 it/sec]
INFO - 16:22:21:     21%|██        | 21/100 [00:00<00:00, 4102.87 it/sec]
INFO - 16:22:21:     22%|██▏       | 22/100 [00:00<00:00, 4113.90 it/sec]
INFO - 16:22:21:     23%|██▎       | 23/100 [00:00<00:00, 4130.37 it/sec]
INFO - 16:22:21:     24%|██▍       | 24/100 [00:00<00:00, 4134.19 it/sec]
INFO - 16:22:21:     25%|██▌       | 25/100 [00:00<00:00, 4150.15 it/sec]
INFO - 16:22:21:     26%|██▌       | 26/100 [00:00<00:00, 4166.74 it/sec]
INFO - 16:22:21:     27%|██▋       | 27/100 [00:00<00:00, 4182.53 it/sec]
INFO - 16:22:21:     28%|██▊       | 28/100 [00:00<00:00, 4183.70 it/sec]
INFO - 16:22:21:     29%|██▉       | 29/100 [00:00<00:00, 4191.85 it/sec]
INFO - 16:22:21:     30%|███       | 30/100 [00:00<00:00, 4203.97 it/sec]
INFO - 16:22:21:     31%|███       | 31/100 [00:00<00:00, 4209.10 it/sec]
INFO - 16:22:21:     32%|███▏      | 32/100 [00:00<00:00, 4221.61 it/sec]
INFO - 16:22:21:     33%|███▎      | 33/100 [00:00<00:00, 4219.75 it/sec]
INFO - 16:22:21:     34%|███▍      | 34/100 [00:00<00:00, 4231.39 it/sec]
INFO - 16:22:21:     35%|███▌      | 35/100 [00:00<00:00, 4243.16 it/sec]
INFO - 16:22:21:     36%|███▌      | 36/100 [00:00<00:00, 4254.22 it/sec]
INFO - 16:22:21:     37%|███▋      | 37/100 [00:00<00:00, 4257.01 it/sec]
INFO - 16:22:21:     38%|███▊      | 38/100 [00:00<00:00, 4265.47 it/sec]
INFO - 16:22:21:     39%|███▉      | 39/100 [00:00<00:00, 4276.21 it/sec]
INFO - 16:22:21:     40%|████      | 40/100 [00:00<00:00, 4286.79 it/sec]
INFO - 16:22:21:     41%|████      | 41/100 [00:00<00:00, 4297.23 it/sec]
INFO - 16:22:21:     42%|████▏     | 42/100 [00:00<00:00, 4279.28 it/sec]
INFO - 16:22:21:     43%|████▎     | 43/100 [00:00<00:00, 4286.00 it/sec]
INFO - 16:22:21:     44%|████▍     | 44/100 [00:00<00:00, 4294.54 it/sec]
INFO - 16:22:21:     45%|████▌     | 45/100 [00:00<00:00, 4303.32 it/sec]
INFO - 16:22:21:     46%|████▌     | 46/100 [00:00<00:00, 4303.67 it/sec]
INFO - 16:22:21:     47%|████▋     | 47/100 [00:00<00:00, 4309.19 it/sec]
INFO - 16:22:21:     48%|████▊     | 48/100 [00:00<00:00, 4316.98 it/sec]
INFO - 16:22:21:     49%|████▉     | 49/100 [00:00<00:00, 4318.48 it/sec]
INFO - 16:22:21:     50%|█████     | 50/100 [00:00<00:00, 4323.85 it/sec]
INFO - 16:22:21:     51%|█████     | 51/100 [00:00<00:00, 4319.66 it/sec]
INFO - 16:22:21:     52%|█████▏    | 52/100 [00:00<00:00, 4293.30 it/sec]
INFO - 16:22:21:     53%|█████▎    | 53/100 [00:00<00:00, 4295.70 it/sec]
INFO - 16:22:21:     54%|█████▍    | 54/100 [00:00<00:00, 4300.95 it/sec]
INFO - 16:22:21:     55%|█████▌    | 55/100 [00:00<00:00, 4299.44 it/sec]
INFO - 16:22:21:     56%|█████▌    | 56/100 [00:00<00:00, 4304.77 it/sec]
INFO - 16:22:21:     57%|█████▋    | 57/100 [00:00<00:00, 4310.54 it/sec]
INFO - 16:22:21:     58%|█████▊    | 58/100 [00:00<00:00, 4315.43 it/sec]
INFO - 16:22:21:     59%|█████▉    | 59/100 [00:00<00:00, 4314.53 it/sec]
INFO - 16:22:21:     60%|██████    | 60/100 [00:00<00:00, 4317.42 it/sec]
INFO - 16:22:21:     61%|██████    | 61/100 [00:00<00:00, 4321.18 it/sec]
INFO - 16:22:21:     62%|██████▏   | 62/100 [00:00<00:00, 4326.25 it/sec]
INFO - 16:22:21:     63%|██████▎   | 63/100 [00:00<00:00, 4331.54 it/sec]
INFO - 16:22:21:     64%|██████▍   | 64/100 [00:00<00:00, 4330.79 it/sec]
INFO - 16:22:21:     65%|██████▌   | 65/100 [00:00<00:00, 4335.03 it/sec]
INFO - 16:22:21:     66%|██████▌   | 66/100 [00:00<00:00, 4338.19 it/sec]
INFO - 16:22:21:     67%|██████▋   | 67/100 [00:00<00:00, 4341.46 it/sec]
INFO - 16:22:21:     68%|██████▊   | 68/100 [00:00<00:00, 4334.93 it/sec]
INFO - 16:22:21:     69%|██████▉   | 69/100 [00:00<00:00, 4335.75 it/sec]
INFO - 16:22:21:     70%|███████   | 70/100 [00:00<00:00, 4340.52 it/sec]
INFO - 16:22:21:     71%|███████   | 71/100 [00:00<00:00, 4345.16 it/sec]
INFO - 16:22:21:     72%|███████▏  | 72/100 [00:00<00:00, 4350.00 it/sec]
INFO - 16:22:21:     73%|███████▎  | 73/100 [00:00<00:00, 4348.16 it/sec]
INFO - 16:22:21:     74%|███████▍  | 74/100 [00:00<00:00, 4352.77 it/sec]
INFO - 16:22:21:     75%|███████▌  | 75/100 [00:00<00:00, 4357.93 it/sec]
INFO - 16:22:21:     76%|███████▌  | 76/100 [00:00<00:00, 4362.73 it/sec]
INFO - 16:22:21:     77%|███████▋  | 77/100 [00:00<00:00, 4363.40 it/sec]
INFO - 16:22:21:     78%|███████▊  | 78/100 [00:00<00:00, 4365.80 it/sec]
INFO - 16:22:21:     79%|███████▉  | 79/100 [00:00<00:00, 4370.39 it/sec]
INFO - 16:22:21:     80%|████████  | 80/100 [00:00<00:00, 4375.28 it/sec]
INFO - 16:22:21:     81%|████████  | 81/100 [00:00<00:00, 4379.60 it/sec]
INFO - 16:22:21:     82%|████████▏ | 82/100 [00:00<00:00, 4379.41 it/sec]
INFO - 16:22:21:     83%|████████▎ | 83/100 [00:00<00:00, 4380.72 it/sec]
INFO - 16:22:21:     84%|████████▍ | 84/100 [00:00<00:00, 4383.85 it/sec]
INFO - 16:22:21:     85%|████████▌ | 85/100 [00:00<00:00, 4385.08 it/sec]
INFO - 16:22:21:     86%|████████▌ | 86/100 [00:00<00:00, 4388.31 it/sec]
INFO - 16:22:21:     87%|████████▋ | 87/100 [00:00<00:00, 4386.08 it/sec]
INFO - 16:22:21:     88%|████████▊ | 88/100 [00:00<00:00, 4389.12 it/sec]
INFO - 16:22:21:     89%|████████▉ | 89/100 [00:00<00:00, 4392.20 it/sec]
INFO - 16:22:21:     90%|█████████ | 90/100 [00:00<00:00, 4395.78 it/sec]
INFO - 16:22:21:     91%|█████████ | 91/100 [00:00<00:00, 4395.23 it/sec]
INFO - 16:22:21:     92%|█████████▏| 92/100 [00:00<00:00, 4397.50 it/sec]
INFO - 16:22:21:     93%|█████████▎| 93/100 [00:00<00:00, 4400.41 it/sec]
INFO - 16:22:21:     94%|█████████▍| 94/100 [00:00<00:00, 4402.44 it/sec]
INFO - 16:22:21:     95%|█████████▌| 95/100 [00:00<00:00, 4405.78 it/sec]
INFO - 16:22:21:     96%|█████████▌| 96/100 [00:00<00:00, 4404.38 it/sec]
INFO - 16:22:21:     97%|█████████▋| 97/100 [00:00<00:00, 4406.55 it/sec]
INFO - 16:22:21:     98%|█████████▊| 98/100 [00:00<00:00, 4408.14 it/sec]
INFO - 16:22:21:     99%|█████████▉| 99/100 [00:00<00:00, 4410.93 it/sec]
INFO - 16:22:21:    100%|██████████| 100/100 [00:00<00:00, 4363.48 it/sec]
INFO - 16:22:21: *** End Sampling execution ***

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 - 16:22:22: 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 0.185 seconds)

Gallery generated by Sphinx-Gallery