Launching Model with API

  

MindSpore SciAI provides users with a high order interface AutoModel, with which the supported model in the model library can be instantiated with one line code.

User can launch training and evaluation process with AutoModel.

Obtaining the Network with AutoModel

User can use the function AutoModel.from_pretrained to get the network models, which are supported in SciAI.

Here we use the model Conservative Physics-Informed Neural Networks (CPINNs) as example. For the codes of CPINNs model, please refer to the link.

The fundamental idea about this model can be found in this paper.

from sciai.model import AutoModel

# obtain the cpinns model.
model = AutoModel.from_pretrained("cpinns")

Training and Fine-tuning with AutoModel

User can use the function AutoModel.train to train the neural networks, and before training, user can use AutoModel.update_config to configure the training parameters or finetune the model by loading the .ckpt file.

The acceptable arguments for API AutoModel.update_config depends on the model instantiated.

from sciai.model import AutoModel

# obtain the cpinns model.
model = AutoModel.from_pretrained("cpinns")
# (optional) load the ckpt file and initialize the model based on the loaded parameters.
model.update_config(load_ckpt=True, load_ckpt_path="./checkpoints/your_file.ckpt", epochs=500)
# train the network with default configuration,
# the figures, data and logs generated will be saved in your execution path.
model.train()

Evaluating with AutoModel

User can evaluate the trained networks with function AutoModel.evaluate.

This function will load the .ckpt files provided in SciAI by default. Alternatively, user can load their own .ckpt file with interface AutoModel.update_config.

from sciai.model import AutoModel

# obtain the cpinns model
model = AutoModel.from_pretrained("cpinns")
# (optional) load the ckpt file provided by user
model.update_config(load_ckpt=True, load_ckpt_path="./checkpoints/your_file.ckpt")
# evaluate the model
model.evaluate()