rmsander

"""Function to compute stability metrics for covariance matrices,
namely log determinants and condition numbers."""
# Use torch
import torch

def compute_covariance_metrics(Kxx, x_train, writer):
    """Helper function to compute covariance metrics.

    Parameters:
        Kxx (torch.Tensor): Tensor object corresponding to a covariance matrix.

rmsander

"""Function for implementing custom distance metrics for use in sklearn neighbor
search algorithms.

Here, we will use a composite product norm composed of a product of norms over
given (contiguous) subsets of states. The norms over these subsets are taken
to be L2 norms.

If x = [s a]^T, where s and a are the contiguous subsets of x that this
product norm is decomposed into, then this product norm can be written as:

rmsander

"""Module for setting a Gym MuJoCo environment's state to a stored state using 
qpos and qvel inputs, which are used for state setting in the MuJoCo simulator.
An example is carried out with the HalfCheetah environment."""

from gym.envs.mujoco.half_cheetah_v3 import HalfCheetahEnv
import numpy as np


def main():
    # Initialize and reset the environment to extract initial state

rmsander

"""Adapted from the Keras VAE guide: https://keras.io/examples/generative/vae/."""
import numpy as np
import tensorflow as tf
from tensorflow import keras
from tensorflow.keras import layers

def make_encoder():
  """Function for making the encoder."""
  latent_dim = 10

rmsander

import numpy as np
import tensorflow as tf

def vector_same_class_sample_from_file(Y, y_sample, base_paths_by_class=None, X=None, num_classes=2):
  """Function for generating equal-class Mixup pairs in a vectorized fashion.
  Essentially, what we'll do is separate X into different sets according to the
  class of the corresponding index in Y. Then, given y_sample, a set of classes
  of the first set of images/vectors that we use for Mixup, we pair the same classes
  together appropriately by only sampling corresponding same-class images/vectors
  from X using appropriate indexing.

rmsander

"""Module for a TF/Keras function for selecting samples of the same class to interpolate (e.g. via
Mixup interpolation)."""

import numpy as np
import tensorflow as tf

def vector_same_class_sample(X, Y, y_sample, num_classes=2):
  """Function for generating equal-class Mixup pairs in a vectorized fashion.
  Essentially, what we'll do is separate X into different sets according to the
  class of the corresponding index in Y. Then, given y_sample, a set of classes

rmsander

import tensorflow as tf
import numpy as np


def mixup_tensorflow(X1, X2, alpha=1.0):
  """Function for implementing Mixup using TensorFlow. 
  Mixup interpolation occurs between corresponding indices 
  of X1[i] and X2[i].
  
  Parameters:

rmsander

If you're using mujoco-py (likely through the MuJoCo suite in gym) on MIT Supercloud, you may encounter issues with POSIX locking if you try to install and setup mujoco-py on the standard shared file system.

To get around this, you can install mujoco-py on a different section (the permanent storage section) of the Supercloud server. Below is a bash script I used to get around issue (you may have to modify some of the commands, such as the conda environment you use):

#!/bin/bash

#SBATCH -c 10
#SBATCH -n 1
#SBATCH --exclusive

rmsander

import tensorflow as tf
import numpy as np


def mixup_tensorflow(X1, X2, alpha=1.0):
  """Function for implementing Mixup using TensorFlow. 
  Mixup interpolation occurs between corresponding indices 
  of X1[i] and X2[i].
  
  Parameters:

rmsander

import gpytorch
import numpy as np
import matplotlib.pyplot as plt
import torch

# Create features
X = torch.tensor(np.arange(100)).float()

# Create targets as noisy function of features
Y = torch.tensor(np.add(np.sin(X / 10), np.random.normal(loc=0, scale=0.5, size=100))).float()