Installation

Note

🚨 This package is still undergoing development. 🚨

It is not recommended use Parallel installation, because it’s somewhat complicated and prepared for the future version for parallel reading of data. Now use Quick installation or Source installation, you can also run MCMTpy in parallel. Please note that the test is performed on macOS Big Sur (11.2.1), so it could be slightly different for other OS.

Dependencies

The package MCMTpy runs on Unix-like systems including Mac and Linux. Package need Python 3.6 or greater, but python 3.9 is not supported for pyfk. It depends on the following Python modules:

  • numpy>=1.14

  • tqdm>=4.19.4

  • matplotlib<=3.1.1

  • mpi4py

  • obspy

  • pyfk

  • pyasdf

  • json5

For parallel:

For data preprocessing and visualization:

We recommend to use anaconda as your python environment, and use conda and pip to install those librarys.

Quick installation

Firstly, make sure Anaconda has been installed, then:

$ conda create -n MCMTpy  python=3.8 numpy=1.16 matplotlib=3.1.1 mpi4py obspy pyasdf json5 tqdm
$ conda activate MCMTpy
$ pip install pyfk
$ pip install MCMTpy

Some errors may occurred of that pyfk not support the new version of cysignals, please:

$ conda uninstall cysignals
$ pip install cysignals==1.10.2
$ pip install pyfk

Successful installation will generate an executable program MCMTpy under the anaconda environment path: /Users/user/opt/anaconda3/bin. And run following code to check the installation:

$ MCMTpy --help

Source installation

$ git clone https://github.com/OUCyf/MCMTpy.git
$ cd MCMTpy
$ python setup.py install

Parallel installation (for future version)

1. Install pyfk

Firstly, make sure Anaconda has been installed, then:

$ conda create -n MCMTpy  python=3.8 numpy=1.16 obspy
$ conda activate MCMTpy
$ pip install pyfk

Some errors may occurred of that pyfk not support the new version of cysignals, please:

$ conda uninstall cysignals
$ pip install cysignals==1.10.2
$ pip install pyfk

🚨 Make sure the new environment does not include the following packages that we will install below: 🚨

2. Install the HDF5 parallel version

  • First, make sure openmpi or other MPI program has been installed, preferably source code installed. You can use which mpicc to get the installation’s path.

  • Activate the MCMTpy environment, and uninstall the mpi4py package in this environment (if it have). Then use pip install the mpi4py package not conda.

$ conda uninstall mpi4py
$ pip install mpi4py

Note

When you use pip install mpi4py, it will use the mpicc in your system. But if you use conda to do the installation, it resolves the dependent environment itself and install conda’s mpicc in your environment which will cause problems for subsequent installations due to the not source code compilation of MPI. After installing mpi4py, you can use which mpicc to check whether it is a system mpicc or a conda mpicc.

  • Then you need to install the parallel version of HDF5. Download the source package, compile and install the parallel version to your path. Your system may already have HDF5 installed. Software such as GMT relies on the HDF5 package, so use h5cc-showconfig to see more information about installation, note that h5cc is usually not shown in parallel. We do not want to uninstall the existing version of HDF5 on our system. This does not conflict with our subsequent installation of the parallel version of HDF5, as long as we specify the installation path of the parallel version.

$ chmod 777 ./configure  # enable read/write access
$ ./configure --enable-parallel --enable-shared --prefix=/your/path/of/hdf5_dir/
$ make
$ make check
$ make install

  • After the installation, the executable program will be generated in ./bin path of your hdf5_dir directory. h5pcc will be generated due to the parallel installation, similar to the h5cc will be generated by the non-parallel installation. Check the installation version information:

$ h5pcc -showconfig

3. Install the h5py parallel version

  • Before installing h5py, you need to uninstall both hdf5 and h5py in your environment, and then download h5py <https://github.com/h5py/h5py> and compile the source code. You need to specify the path of HDF5 and CC.

$ export CC=/path/to/mpicc   # do not use conda's mpicc
$ echo %CC
$ export HDF5_MPI="ON"
$ export HDF5_DIR="/your/path/of/hdf5_dir/"
$ pip install .

  • Check the installation version information (download the demo2.py on h5py):

#!/usr/bin/env python3
# -*- coding: utf-8 -*-

from mpi4py import MPI
import h5py

rank = MPI.COMM_WORLD.rank  # The process ID (integer 0-3 for 4-process run)

f = h5py.File('parallel_test.hdf5', 'w', driver='mpio', comm=MPI.COMM_WORLD)

dset = f.create_dataset('test', (4,), dtype='i')
dset[rank] = rank

f.close()

Run the program:

$ mpiexec -n 4 python demo2.py

Looking at the file with h5dump:

$ h5dump parallel_test.hdf5
HDF5 "parallel_test.hdf5" {
GROUP "/" {
   DATASET "test" {
      DATATYPE  H5T_STD_I32LE
      DATASPACE  SIMPLE { ( 4 ) / ( 4 ) }
      DATA {
      (0): 0, 1, 2, 3
      }
   }
}
}

1. Install pyasdf parallel version

  • After the above steps are completed sucessful, you can directly use pip to install pyasdf

$ pip install pyasdf
$ python -c "import pyasdf; pyasdf.print_sys_info()"   # (check for a parallel version)

5. Install MCMTpy parallel version

  • Finally, you can install the parallel version of MCMTpy.

$ pip install MCMTpy        # or python setup.py install
$ pip uninstall MCMTpy      # uninstall

Reference Installation Blog