https://github.com/jakeret/robotics

Sequence Modelling for Position Tracking of Soft Robotic Arm using Temporal Convolutional Networks
https://github.com/jakeret/robotics

Last synced: about 1 month ago
JSON representation

Sequence Modelling for Position Tracking of Soft Robotic Arm using Temporal Convolutional Networks

• Host: GitHub
• URL: https://github.com/jakeret/robotics
• Owner: jakeret
• Created: 2020-05-19T07:51:34.000Z (over 4 years ago)
• Default Branch: master
• Last Pushed: 2022-12-26T21:34:46.000Z (about 2 years ago)
• Last Synced: 2023-02-26T07:03:02.525Z (almost 2 years ago)
• Language: Jupyter Notebook
• Size: 787 KB
• Stars: 0
• Watchers: 1
• Forks: 0
• Open Issues: 13
• Metadata Files:
  • Readme: README.md

Awesome Lists containing this project

README

        
Sequence Modelling for Position Tracking of Soft Robotic Arm using Temporal Convolutional Networks

====

Setup

-----

Setup virtualenv and dependencies by running:

```

$ pipenv install --dev 

$ pipenv shell

```

Training

--------

A single model can be trained by using `robotics.py`, either in a scripts using

```

import robotics

robotics.run_training(data_path, log_path, output_path, **hyperparams)

``` 

or on the command line with

```

$ python robotics.py --help

usage: robotics.py [-h] [--data-path DATA_PATH] [--output-path OUTPUT_PATH]

                   [--log-path LOG_PATH] [--depth DEPTH]

                   [--kernel_size KERNEL_SIZE] [--filters FILTERS]

                   [--dropout_rate DROPOUT_RATE]

                   [--normalize_inputs NORMALIZE_INPUTS]

                   [--learning_rate LEARNING_RATE] [--batch_size BATCH_SIZE]

                   [--epochs EPOCHS]

optional arguments:

  -h, --help            show this help message and exit

  --data-path DATA_PATH

  --output-path OUTPUT_PATH

  --log-path LOG_PATH

  --depth DEPTH

  --kernel_size KERNEL_SIZE

  --filters FILTERS

  --dropout_rate DROPOUT_RATE

  --normalize_inputs NORMALIZE_INPUTS

  --learning_rate LEARNING_RATE

  --batch_size BATCH_SIZE

  --epochs EPOCHS

```

Parameter tuning

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

A hyperparameter grid search can be starter using `tuning.py`, either in a scripts using

```

import tuning

tuning.tune_model(data_path="", log_path="logs", output_path="model")

``` 

or on the command line with

```

python tuning.py --help

usage: tuning.py [-h] [--data-path DATA_PATH] [--output-path OUTPUT_PATH]

                 [--log-path LOG_PATH] [--learning_rate LEARNING_RATE]

                 [--batch_size BATCH_SIZE] [--epochs EPOCHS]

optional arguments:

  -h, --help            show this help message and exit

  --data-path DATA_PATH

  --output-path OUTPUT_PATH

  --log-path LOG_PATH

  --learning_rate LEARNING_RATE

  --batch_size BATCH_SIZE

  --epochs EPOCHS

```

Evaluation

----------

All training results are being tracked with mlflow and TensorBoard. To launch the UI's run

```

$ pipenv shell

$ mlflow ui

```

```

$ pipenv shell

$ tensorboard --logdir=./logs

```