enggiqbal/MPSE

Multi-Perspective, Simultaneous Embedding (MPSE) interactive system

MPSE: with webserver (optional)

1. Install NodeJS 10.16.0. (download from nodejs website)
2. Run npm install in this directory to install the dependencies.
3. Run npm run build to build the static files.
4. Run nodemon server.js to start the server

list of dependencies at python3.6

matplotlib==2.2.2
networkx==2.1
pandas==0.23.3
numpy==1.14.5
scipy==1.1.0
autograd==1.2
torch==1.0.1.post2
scikit_learn==0.21.3

pip3.6 install -r requirements.txt

MPSE: Jupyter Notebook examples

MPSE/mview_examples

MPSE: python command line

usage: mpse.py [-h] -d D [D ...] [-o OUTPUT_DIR] [-e EXPERIMENT_NAME]
               [-max_iters MAX_ITERS] [-n SAMPLE_SIZE] [-X0 {True,False}]
               [-ps {fixed,same,standard,cylinder,orthogonal,normal,uniform,variable}]
               [-vt {pointbased,attributebased}] [-an AVERAGE_NEIGHBORS]
               [-ds PRELOADED_DATASET]

MPSE

optional arguments:
  -h, --help            show this help message and exit
  -d D [D ...], --d D [D ...]
                        List of input files with distace matices
  -o OUTPUT_DIR, --output_dir OUTPUT_DIR
                        Output directory
  -e EXPERIMENT_NAME, --experiment_name EXPERIMENT_NAME
                        Experiment name
  -max_iters MAX_ITERS, --max_iters MAX_ITERS
                        Max iterations
  -n SAMPLE_SIZE, --sample_size SAMPLE_SIZE
                        Number of samples
  -X0 {True,False}, --X0 {True,False}
                        Smart initialization
  -ps {fixed,same,standard,cylinder,orthogonal,normal,uniform,variable}, --projection_type {fixed,same,standard,cylinder,orthogonal,normal,uniform,variable}
                        projection set
  -vt {pointbased,attributebased}, --visualization_template {pointbased,attributebased}
                        Visualization template
  -an AVERAGE_NEIGHBORS, --average_neighbors AVERAGE_NEIGHBORS
                        average neighbors
  -ds PRELOADED_DATASET, --preloaded_dataset PRELOADED_DATASET
                        Preloaded Dataset

examples of command

pass distance matrices

python3 mpse.py -d MPSE/datasets/dataset_tabluar/data/dissimple1000_1.csv  MPSE/datasets/dataset_tabluar/data/dissimple1000_2.csv  MPSE/datasets/dataset_tabluar/data/dissimple1000_3.csv

run circlesquare example with 150 points, maximum iteration 100 then save output to mytest directory.

python3.6 mpse.py -ds circlesquare -n 150 -max_iters 100 -e mytest

iqbal@on-campus-10-138-77-23 MPSE-web % python3.6 mpse.py -ds circlesquare -n 150 -max_iters 100 -e mytest
<h1>Please keep the window running</h1>
Total Samples found:100<br>
mpse.MPSE():
multigraph.DISS():
  nodes : 100
  added attribute:
    type : matrix
    complete : True
  added attribute:
    type : matrix
    complete : True
  MPSE.initialize():
    X0 : random
    Q0 : given
  dissimilarity stats:
    number of views : 2
    number of points : 100
  embedding stats:
    embedding dimension : 3
    projection dimension : 2
  MPSE.gd():
    mpse method : fixed projections
    initial stress : 5.89e-01
gd.single():
  computation parameters:
    stochastic : True
    constraint : False
    scheme : mm
    initial lr : 1
    min_cost : 1.00e-03
    max_iter : 100
    max_step : 1.00e+10
  progress:
      99/100 : cost = 4.31e-02, grad = 4.97e-03, lr = 2.20e+00, step = 1.09e-02
  results:
    conclusion : maximum number of iterations reached
    total iterations : 99
    final cost : 4.31e-02
    final gradient size : 4.97e-03
    final learning rate : 2.20e+00
    final step size : 1.09e-02
    time : 1.37e+01 [sec]
  Final stress : 4.27e-02
Saving js data in: MPSE/outputs/mytest/coordinates.js
JS file was saved in:  MPSE/outputs/mytest/coordinates.js
DEPRECATION WARNING: The system version of Tk is deprecated and may be removed in a future release. Please don't rely on it. Set TK_SILENCE_DEPRECATION=1 to suppress this warning.
**<br> output path: MPSE/outputs/mytest/index.html**
<br><h1> <a target='_blank'  href ='static/mytest/index.html'>Interactive visualization</a></h1><br>
<br><h2> <a target='_blank'  href ='static/mytest/mytest_pos.csv'>Output 3D position was saved here</a></h2><br>
<br><h2> <a target='_blank'  href ='static/mytest/coordinates.js'>Output details (history, projections, position) was saved here</a></h2><br>
<br>cost history saved as  MPSE/outputs/mytest/cost.png
<br><img src=/static/mytest/cost.png>

python3.6 mpse.py -ds 123 -max_iters 500 -ps cylinder -e 123 -an 4

Citation (draft version )

@misc{hossain2019multiperspective,
    title={Multi-Perspective, Simultaneous Embedding},
    author={Md Iqbal Hossain and Vahan Huroyan and Stephen Kobourov and Raymundo Navarrete},
    year={2019},
    eprint={1909.06485},
    archivePrefix={arXiv},
    primaryClass={cs.DS}
}