### Training Example

Source: https://github.com/mindspore-ai/mindscience/blob/master/MindChem/applications/nequip/README.md

Command to start the training process for the NequIP model, specifying configuration file, execution mode, target device, device ID, and data type.

```bash
python train.py --config_file_path ./rmd.yaml --mode GRAPH --device_target Ascend --device_id 0 --dtype float32
```

--------------------------------

### Dataset Structure Example

Source: https://github.com/mindspore-ai/mindscience/blob/master/MindChem/applications/crystalflow/README.md

Example of how the dataset folder should be structured.

```txt
crystalflow
    ... 
    └─dataset
            perov_5 钙钛矿数据集
            carbon_24 碳晶体数据集
            mp_20 晶胞内原子数最多为20的MP数据集
            mpts_52 晶胞内原子数最多为52的MP数据集
            dataset_prop.txt 数据集属性文件
    ...

```

--------------------------------

### Install MindSpore

Source: https://github.com/mindspore-ai/mindscience/blob/master/MindEnergy/applications/DeepONet-Grid-UQ/README.md

Command to install the MindSpore framework.

```bash
pip install mindspore

```

--------------------------------

### Python Training Command

Source: https://github.com/mindspore-ai/mindscience/blob/master/MindChem/applications/diffcsp/README.md

Command to install requirements and start training.

```bash
pip install -r requirement.txt
python train.py
```

--------------------------------

### Start training

Source: https://github.com/mindspore-ai/mindscience/blob/master/MindEarth/applications/sea/LeadFormer/LeadFormer.ipynb

Initializes the trainer with the configuration and starts the training process.

```python
epoch_size = config["train"].get("epochs", 300)
trainer = Trainer(config, epochs=epoch_size)
trainer.train()
```

--------------------------------

### Unzip Example PDBs

Source: https://github.com/mindspore-ai/mindscience/blob/master/MindSPONGE/applications/proteinmpnn/README.md

Command to unzip example PDB files for running demonstrations.

```bash
unzip examples/example_inputs.zip -d examples/
```

--------------------------------

### Extract Example Data

Source: https://github.com/mindspore-ai/mindscience/blob/master/MindSPONGE/applications/rf_diffusion/README.md

Commands to extract example PDB and scaffold files for running demonstrations.

```bash
tar -xvf examples/ppi_scaffolds_subset.tar.gz -C examples/
tar -xvf examples/antibody_pdbs.tar.gz -C examples/
tar -xvf examples/input_pdbs.tar.gz -C examples/
tar -xvf examples/target_folds.tar.gz -C examples/
tar -xvf examples/tim_barrel_scaffold.tar.gz -C examples/
```

--------------------------------

### Modular Installation

Source: https://github.com/mindspore-ai/mindscience/blob/master/docs/architecture/mindscience-core.md

Configuration for setup.py to support on-demand installation of extras like 'mindflow', 'mindsponge', or 'all'.

```python
# 支持按需安装
extras_require={
    "mindflow": ["matplotlib", "h5py"],
    "mindsponge": ["biopython"],
    "all": ["matplotlib", "h5py", "biopython"]
}

```

--------------------------------

### Install Dependencies

Source: https://github.com/mindspore-ai/mindscience/blob/master/MindChem/applications/orb/orb.ipynb

Installs the required packages listed in the requirement.txt file.

```python
!pip install -r requirement.txt
```

--------------------------------

### Install Dependencies

Source: https://github.com/mindspore-ai/mindscience/blob/master/MindEnergy/applications/PowerFlowNet/README_MINDSPORE_MIGRATION.md

Commands to install MindSpore (CPU/GPU/Ascend) and other necessary Python packages.

```bash
# MindSpore (CPU/GPU)
pip install mindspore

# MindSpore Ascend (in Huawei Cloud environment)
# Follow official documentation: https://www.mindspore.cn/install

# Other dependencies
pip install numpy torch-geometric h5py matplotlib
```

--------------------------------

### Training Log Example

Source: https://github.com/mindspore-ai/mindscience/blob/master/MindChem/applications/nequip/README_CN.md

Example log output during the training process.

```log
2024-03-25 21:49:49 (INFO): ---- Configuration Summary -----
.
.
.
2024-03-25 21:49:49 (INFO): --------------------------------
2024-03-25 21:49:49 (INFO): Loading data...
2024-03-25 21:50:13 (INFO): Initializing model...
2024-03-25 21:50:37 (INFO): Initializing train...
2024-03-25 22:01:58 (INFO): epoch 1:  train loss: 1000.02729235, time gap: 680.55, total time used: 680.55
.
.
.

```

--------------------------------

### Auxiliary Potential Function Example

Source: https://github.com/mindspore-ai/mindscience/blob/master/MindSPONGE/applications/rf_diffusion/README.md

Example command for setting guiding potentials, intra/inter weights, guide scale, and decay.

```bash
potentials.guiding_potentials=["type:olig_contacts,weight_intra:1,weight_inter:0.1"] potentials.olig_intra_all=True potentials.olig_inter_all=True potentials.guide_scale=2 potentials.guide_decay='quadratic'
```

--------------------------------

### Create Trainer and Training Configuration

Source: https://github.com/mindspore-ai/mindscience/blob/master/MindEnergy/applications/DeepONet-Grid-UQ/DeepONet_Grid_UQ_Probabilistic.ipynb

Initializes the trainer and sets up the training configuration.

```python
# ===================================
# Create trainer and training configuration
# ===================================
print("\n" + "=" * 50)
print("Creating trainer and training configuration")
print("=" * 50)


```

--------------------------------

### Complex Design with Hotspot Residues

Source: https://github.com/mindspore-ai/mindscience/blob/master/MindSPONGE/applications/rf_diffusion/README.md

Example of specifying hotspot residues for guiding complex design.

```bash
'ppi.hotspot_res=[A30,A33,A34]'
```

--------------------------------

### PSSM-Guided Design Example

Source: https://github.com/mindspore-ai/mindscience/blob/master/MindSPONGE/applications/proteinmpnn/README.md

Bash script to mix ProteinMPNN logits with PSSM probabilities for guided design.

```bash
bash examples/submit_example_pssm.sh
```

--------------------------------

### Command-line Argument Parsing for Case Selection

Source: https://github.com/mindspore-ai/mindscience/blob/master/CONTRIBUTION.md

Example of how to specify the case to run using command-line arguments.

```python
import os
import time
import argparse
import numpy as np

from mindspore import context, nn, Tensor, set_seed, ops, data_sink, jit, save_checkpoint
from mindspore import dtype as mstype

from mindflow import FNO1D, load_yaml_config, get_warmup_cosine_annealing_lr
from mindflow.pde import FlowWithLoss

from src import create_training_dataset, visual, calculate_l2_error
# Relevant dependency imports, in the order of Python official libraries, third-party libraries, mindflow, src

set_seed(123456)
np.random.seed(123456)
```

--------------------------------

### Directory Structure for a Single Case

Source: https://github.com/mindspore-ai/mindscience/blob/master/CONTRIBUTION.md

Example directory structure for a single case, such as PINNs solving Burgers equation.

```shell
.
├── images
│  ├── background.png
│  └── result.png
├── src
│  ├── __init__.py
│  ├── dataset.py
│  ├── model.py
│  └── utils.py
├── configs
│  ├── fno1d.yaml
├── README.md
├── README_CN.md
├── problem.ipynb
├── problem_CN.ipynb
├── burgers_cfg.yaml
├── eval.py
└── train.py
```

--------------------------------

### Loss Function and Optimizer setup

Source: https://github.com/mindspore-ai/mindscience/blob/master/MindEnergy/applications/DAE-PINN/DAE-PINN_EN.ipynb

Initializes the DaeTrainer, loads IRK weights and times, sets up the Adam optimizer, and configures a learning rate scheduler.

```Python
num_irk_stages = model_params['num_irk_stages']
# collecting RK data
data_dir = data_params['data_dir']
irk_data = np.loadtxt(os.path.join(data_dir, 'IRK_weights', f'Butcher_IRK{num_irk_stages}.txt'), ndmin=2, dtype=np.float32)
irk_weights = np.reshape(irk_data[0:num_irk_stages**2+num_irk_stages], (num_irk_stages+1, num_irk_stages))
irk_weights = Tensor(irk_weights)
irk_times = irk_data[num_irk_stages**2 + num_irk_stages:]
trainer = DaeTrainer(net, irk_weights=irk_weights, irk_times=irk_times, h=ode_params['h'],
                     dyn_weight=model_params['dyn_weight'], alg_weight=model_params['alg_weight'])

optimizer = optim.Adam(net.trainable_params(), lr=float(optim_params['lr']))
scheduler_type = optim_params['scheduler_type']
use_scheduler = optim_params['use_scheduler']
if use_scheduler:
    if scheduler_type == "plateau":
        scheduler = optim.lr_scheduler.ReduceLROnPlateau(
            optimizer,
            mode='min',
            patience=optim_params['patience'],
            factor=optim_params['factor'],
        )
    elif scheduler_type == "step":
        scheduler = optim.lr_scheduler.StepLR(
            optimizer, step_size=optim_params['patience'], gamma=optim_params['factor']
        )
    else:
        scheduler = None
else:
    scheduler = None
```

--------------------------------

### Directory Structure for Multiple Cases

Source: https://github.com/mindspore-ai/mindscience/blob/master/CONTRIBUTION.md

Example directory structure for multiple cases that share the same model and method but use different datasets.

```shell
.
├──images
│  ├──background.png
│  ├──result1.png
│  ├──result2.png
│  └──result3.png
├──src
│  ├──__init__.py
│  ├──dataset.py
│  ├──model.py
│  └──utils.py
├──configs
│  ├──fno1d.yaml
├──README.md
├──README_CN.md
├──problem.ipynb
├──problem_CN.ipynb
├──problem_cfg.yaml
├──eval.py
└──train.py
```

--------------------------------

### API Reference - Device Initialization

Source: https://github.com/mindspore-ai/mindscience/blob/master/MindEnergy/applications/PowerFlowNet/README_MINDSPORE_MIGRATION.md

Code snippets for initializing the device (CPU, GPU, Ascend) and getting device configuration.

```python
from powerflownet.configs import init_device, DeviceConfig

# Initialize device
init_device('Ascend')  # 'CPU', 'GPU', 'Ascend'

# Get device configuration
config = DeviceConfig()
device = config.get_device_target()
```

--------------------------------

### Import and Usage Example

Source: https://github.com/mindspore-ai/mindscience/blob/master/MindEnergy/applications/PowerFlowNet/README_MINDSPORE_MIGRATION.md

Python code demonstrating how to import models, load data, and set up a training loop using MindSpore.

```python
from powerflownet.src import MLPNet, MPN, GCNNet
from powerflownet.src import PowerFlowDataV2, PowerFlowDataLoaderV2
from powerflownet.configs import init_device

# Initialize device
init_device('Ascend')  # or 'CPU', 'GPU'

# Create model
model = MPN(
    nfeature_dim=4,
    efeature_dim=2,
    output_dim=4,
    hidden_dim=64,
    n_gnn_layers=3,
    k=3
)

# Load data
dataset = PowerFlowDataV2(root='./data', case='14v2', split=[0.7, 0.15, 0.15], task='train')
dataloader = PowerFlowDataLoaderV2(dataset, batch_size=256, shuffle=True)

# Train
import mindspore as ms
from mindspore import nn

optimizer = nn.Adam(model.trainable_params(), learning_rate=0.001)
loss_fn = MaskedL2Loss(regularize=False)

def forward_fn(batch):
    pred = model(batch)
    loss = loss_fn(pred, batch.y, batch.pred_mask)
    return loss

grad_fn = ms.value_and_grad(forward_fn, None, model.trainable_params())

for batch in dataloader:
    loss, grads = grad_fn(batch)
    optimizer(grads)
```

--------------------------------

### Symmetric Motif Scaffolding Example Script

Source: https://github.com/mindspore-ai/mindscience/blob/master/MindSPONGE/applications/rf_diffusion/README_en.md

This script can be used to scaffold C4 symmetric Nickel binding domains, combining symmetric oligomer generation, motif scaffolding, and guiding potentials.

```bash
examples/design_nickel.sh
```

--------------------------------

### Train with Real Data

Source: https://github.com/mindspore-ai/mindscience/blob/master/MindEnergy/applications/DeepONet-Grid-UQ/README.md

Command to start training using the provided dataset.

```bash
python train.py

```

--------------------------------

### FAQ - Running on Ascend

Source: https://github.com/mindspore-ai/mindscience/blob/master/MindEnergy/applications/PowerFlowNet/README_MINDSPORE_MIGRATION.md

Code snippet to address the issue of the first epoch taking a long time on Ascend devices.

```python
if 'RANK_TABLE_FILE' in os.environ:
    del os.environ['RANK_TABLE_FILE']
import mindspore as ms
ms.set_context(mode=ms.PYNATIVE_MODE)
```

--------------------------------

### Load configuration and initialize trainer

Source: https://github.com/mindspore-ai/mindscience/blob/master/MindEarth/applications/sea/LeadFormer/LeadFormer.ipynb

Loads the YAML configuration file and initializes the Trainer object.

```python
from mindearth.utils import load_yaml_config

from src.solver import Trainer
from src.forecast import Tester
```

--------------------------------

### Install Dependencies

Source: https://github.com/mindspore-ai/mindscience/blob/master/MindChem/applications/crystalflow/README.md

Installs the required Python packages for the project.

```bash
pip install -r requirement.txt
```

--------------------------------

### Training Configuration and Trainer Creation

Source: https://github.com/mindspore-ai/mindscience/blob/master/MindEnergy/applications/DeepONet-Grid-UQ/DeepONet_Grid_UQ_Probabilistic.ipynb

Defines the configuration dictionary for training the model and then creates the trainer instance.

```python
config = {
    'model': {
        'm': m,
        'dim': dim,
        'width': width,
        'depth': depth,
        'n_basis': n_basis,
        'branch_type': branch_type,
        'trunk_type': trunk_type,
        'activation': activation,
        'use_bias': True
    },
    'training': {
        'batch_size': batch_size,
        'epochs': n_epochs,
        'log_interval': 100,
        'eval_interval': 1000,
        'scheduler': scheduler,
        'min_lr': min_lr,
        'max_lr': max_lr,
        'total_step': total_step,
        'step_per_epoch': step_per_epoch,
        'decay_epoch': decay_epoch,
        'verbose': verbose,
        'loss_type': 'nll',
        'optimizer': 'adam',
        'weight_decay': 0.0
    },
    'output': {
        'save_dir': 'outputs',
        'save_best': True
    }
}

# Create trainer
trainer = create_trainer(model, config, save_dir='outputs', distributed=0)
print("Trainer created successfully")
```

--------------------------------

### Evaluation Result Example

Source: https://github.com/mindspore-ai/mindscience/blob/master/MindChem/applications/crystalflow/README.md

Example of the JSON output for evaluation metrics.

```json
{"match_rate": 0.6107671899181959, "rms_dist": 0.07492558322002925}

```

--------------------------------

### Quick Start - Training

Source: https://github.com/mindspore-ai/mindscience/blob/master/MindChem/applications/orb/README.md

Commands for initiating single-card and multi-card training of the Orb model.

```bash
bash run.sh
```

```bash
bash run_parallel.sh
```

--------------------------------

### Example Evaluation Output

Source: https://github.com/mindspore-ai/mindscience/blob/master/MindChem/applications/crystalflow/README_EN.md

Example JSON output from the evaluation process.

```json
{"match_rate": 0.6107671899181959, "rms_dist": 0.07492558322002925}
```

--------------------------------

### Evaluation Result Example

Source: https://github.com/mindspore-ai/mindscience/blob/master/MindChem/applications/diffcsp/README.md

Example JSON output for evaluation results.

```json
{"match_rate": 0.985997357992074, "rms_dist": 0.013073775170360118}
```

--------------------------------

### Start Monitoring Service

Source: https://github.com/mindspore-ai/mindscience/blob/master/MindEnergy/applications/deploy/README.md

Command to start the monitoring service.

```bash
python monitor.py
```

--------------------------------

### Project Root Setup

Source: https://github.com/mindspore-ai/mindscience/blob/master/MindFlow/applications/data_driven/burgers/fno1d/FNO1D.ipynb

Appends the project root directory to sys.path for module imports.

```python
import sys
import os
project_root = os.path.abspath("../../../../..")
sys.path.append(project_root)
```

--------------------------------

### Start Deployment Service

Source: https://github.com/mindspore-ai/mindscience/blob/master/MindEnergy/applications/deploy/README.md

Command to start the deployment service.

```bash
python deploy.py
```

--------------------------------

### Progress Output Example

Source: https://github.com/mindspore-ai/mindscience/blob/master/MindEarth/applications/sea/LeadFormer/LeadFormer.ipynb

Example of progress output during model execution, showing iteration progress and potential runtime warnings.

```text
100%|█████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 1301/1301 [11:51<00:00,  1.83it/s]
/tmp/ipykernel_1012809/934503037.py:629: RuntimeWarning: All-NaN slice encountered
  disnrst = np.nanmin(dismin, axis=0)
```

--------------------------------

### Load Model

Source: https://github.com/mindspore-ai/mindscience/blob/master/MindEnergy/applications/deploy/EXAMPLE.md

Use curl to load a model.

```bash
curl -X POST "http://localhost:8001/mindscience/deploy/load_model" \
  -H "Content-Type: multipart/form-data" \
  -F "model_name=your_model" \
  -F "model_file=@/path/to/your/model_file.zip"
```

--------------------------------

### Install HMMER

Source: https://github.com/mindspore-ai/mindscience/blob/master/MindSPONGE/applications/AlphaFold3/README.md

Download, compile, and install HMMER, then add its bin directory to the PATH.

```bash
mkdir /path/to/hmmer_build /path/to/hmmer && \
mv ./hmmer-3.4.tar.gz /path/to/hmmer_build && \
cd /path/to/hmmer_build && tar -zxf hmmer-3.4.tar.gz && rm hmmer-3.4.tar.gz && \
cd /path/to/hmmer_build/hmmer-3.4 && ./configure --prefix=/path/to/hmmer && \
make -j8 && make install && \
cd /path/to/hmmer_build/hmmer-3.4/easel && make install && \
rm -rf /path/to/hmmer_build
export PATH=/hmmer/bin:$PATH
which jackhmmer
```

--------------------------------

### Install Dependencies

Source: https://github.com/mindspore-ai/mindscience/blob/master/MindSPONGE/applications/AlphaFold3/README.md

Install the required Python packages and set the PYTHONPATH environment variable.

```bash
pip install -r requirements.txt
#`{PATH}` is the current directory
export PYTHONPATH={PATH}/mindscience
```

--------------------------------

### Model File Structure

Source: https://github.com/mindspore-ai/mindscience/blob/master/MindEnergy/applications/deploy/EXAMPLE.md

The directory format for model_file.zip.

```bash
model_file.zip
├── your_model_1.mindir
├── your_model_2.mindir
├── ...
└── your_model_n.mindir
```

--------------------------------

### Install MindScience

Source: https://github.com/mindspore-ai/mindscience/blob/master/MindEnergy/applications/DAE-PINN/DAE-PINN.ipynb

Code snippet to install the MindScience library, with options for Ascend backend.

```python
mindscience_version = "0.1.0"  # update if needed
# Comment out the following code if you are using NPU.
!pip uninstall -y mindscience-ascend
!pip install mindscience-ascend==$mindscience_version

# NPU Uncomment if needed.
# !pip uninstall -y mindscience-ascend
# !pip install mindscience-ascend==$mindscience_version
```

--------------------------------

### Download Results

Source: https://github.com/mindspore-ai/mindscience/blob/master/MindEnergy/applications/deploy/EXAMPLE.md

Download the results of a completed inference task. Replace {task_id} with the actual task ID.

```bash
curl -X GET "http://localhost:8001/mindscience/deploy/query_results/{task_id}" -o "results.h5"
```

```bash
curl -X GET "http://localhost:8001/mindscience/deploy/query_results/123e4567-e89b-12d3-a456-426614174000" -o "results.h5"
```

--------------------------------

### RFantibody Fv Design Example

Source: https://github.com/mindspore-ai/mindscience/blob/master/MindSPONGE/applications/rf_diffusion/README.md

Example configuration for designing an antibody Fv fragment.

```bash
python run_inference.py \
    --config-name antibody \
    antibody.target_pdb=./examples/antibody_pdbs/rsv_site3.pdb \
    antibody.framework_pdb=./examples/antibody_pdbs/hu-4D5-8_Fv.pdb \
    inference.ckpt_override_path=./models/RFdiffusion_Ab.ckpt \
    'ppi.hotspot_res=[T305,T456]' \
    'antibody.design_loops=[L1:8-13,L2:7,L3:9-11,H1:7,H2:6,H3:5-13]' \
    inference.num_designs=2 \
    diffuser.T=50 \
    inference.deterministic=True \
    inference.output_prefix=example_outputs/ab_des
```

--------------------------------

### Dataset Directory Structure

Source: https://github.com/mindspore-ai/mindscience/blob/master/MindChem/applications/diffcsp/README.md

Example structure for the dataset folder.

```txt
diffcsp
    ...
    └─dataset
            perov_5 钙钛矿数据集
            carbon_24 碳晶体数据集
            mp_20 晶胞内原子数最多为20的MP数据集
            mpts_52 晶胞内原子数最多为52的MP数据集
            dataset_prop.txt 数据集属性文件
    ...
```

--------------------------------

### Multi-card Training

Source: https://github.com/mindspore-ai/mindscience/blob/master/MindEnergy/applications/DeepONet-Grid-UQ/README.md

Command for multi-card training.

```bash
msrun --worker_num 8 --local_worker_num 8 --log_dir msrun_log python train.py --distributed 1

```

--------------------------------

### Install Python Dependencies

Source: https://github.com/mindspore-ai/mindscience/blob/master/MindEnergy/applications/deploy/README.md

Command to install Python dependencies from the requirements.txt file.

```bash
pip install -r requirements.txt
```