Based on the artifical Neural Network to fit the optical parameters
Before starting this project, you should install following libraries:
- Pytorch cpu or cuda version are both fine.
- Numpy; matplotlib
You can modify the training details/saving path in config.py.
You can start to train model by following command: python train.py.
The function of each py.files are defined as follows:
-
config.py
vth： the nth velocity with range of v1~v6
epoch： total step of training process.
N1/N2 mean that the nums of nerons in each layer and layesr is the nums of hidden layers.
save_path is the training saving path, which include the training results.
save_infer_path include the inference results.
-
train.py
Your can modify the training parameters in config.py. For example, your can change the vth to modify the target vth velocity in the machine learning process.
-
inference.py
(1). inference_results_mesh().
In this function, you can infer the vth by feeding the mesh inputs into the well training model. The process of this calculation function is :
for x in (-0.05, 0.05, 0.01):
for y in (-0.05, 0.05, 0.01):
for z in (-0.05, 0.05, 0.01):
vth = model([x, y, z])(2). inference_results_uniform().
In this function, you can infer the vth by feeding the uniform inputs into the well training model. The process of this calculation function is:
for x in (-0.05, 0.05, 0.01):
y, z = x
vth = model([x, y, z])Additionally, you can change the strain direction by modifying the input_d vector. For example, input_d = [1, 0, 0] corresponds to strain along the x-direction, [1, 1, 0] corresponds xy direction.