Forward modeling

Seistorch now using mpi4py library for performing forward modeling (# WILL BE DEPRECATED, USING TORCHRUN INSTEAD). The following files are needed for forward modeling:

host file

The host file is used to specify the nodes and gpus u wanna use. It has following format:
```
ip address:number of processes
# The number of processes must be <= gpu device counts
```
Example:
```
192.168.0.1:5
192.168.0.2:4
192.168.0.2:4
```
For the example case, 4 gpu cards per node is used, the reason why the value of the first node reads 5 is that a extra rank is needed for task assignment.
source and receiver files (in pkl format)

Please refer to Data format.
configure file (in yml format)

Please refer to example configure files in examples/forward_modeling/different_eq/configs/*.yml

executable file

An executable forward.sh file looks like:

mpirun -f hosts \
python fwi.py configure.yml \
--mode forward \
--num-batches 1 \
--use-cuda

2d forward modeling in acoustic case

The code of this section locates at examples/forward_modeling/forward_modeling2d. This example shows how to run forward modeling with your own model and geometry.

First we need to change the directory into it.
```
cd examples/forward_modeling
```
Generate a two layer model and the corresponding sources-receivers pairs by typing:
```
python generate_model_geometry.py
```
Two new folders geometry and velocity_model will be created. The figure model_geometry.png illustrates the generated layer model and the locations of source-receiver pairs.
Running the shell script forward.sh, a file called shot_gather.npy will be created.
```
sh forward.sh
```
Show the shot gathers.
```
python show_shotgather.py
```
The plotted results will be save in shot_gather.png (shown as follows).

2d forward modeling in elastic case

The code of this section locates at examples/forward_modeling/elastic_modeling2d. This example shows how to run forward modeling using elastic wave equation.

Step1: Generate a two layer model and the corresponding sources-receivers pairs by typing:
```
python generate_model_geometry.py
```
Two new folders geometry and velocity_model will be created. The figure model_geometry.png illustrates the generated layer model and the locations of source-receiver pairs.
Step2: Running the shell script forward.sh, a file called shot_gather.npy will be created.
```
sh forward.sh
```
Step3: Show the shot gathers.
```
python show_shotgather.py
```
The plotted results will be save in shot_gather.png (shown as follows).

3d forward modeling in acoustic case

The code of this section locates at examples/forward_modeling/forward_modeling3d. This example shows how to run forward modeling with your own model and geometry.

Generate geometry and velocity model

The script generate_model_geometry.py generates a 3D velocity model with two layers. A ricker source at the center of the model suface is used for modeling. Moreover, we have created a three-dimensional observational system, and a schematic diagram of this observational system will be generated after running this script.
```
python generate_model_geometry.py
```
Run forward modeling

Perform simulation by running script forward.sh.
```
sh forward.sh
```
Show results

The modeled data has 1 shot with 2000 time samples, 128 traces and a single component (displacement in scalar wave equation). The first 64 and last 64 traces are recorded along different line directions. Run the script will show the recorded data.
```
python show_shotgather.py
```

If you wanna generate your own 3D geometry and 3D velocity model, please refer to the section data format.