Function unNormalizeData

deterministic/utils/data_utils.py:128–164

Borrowed from SRNN code. Reads a csv file and returns a float32 matrix. https://github.com/asheshjain399/RNNexp/blob/srnn/structural_rnn/CRFProblems/H3.6m/generateMotionData.py#L12 Args normalizedData: nxd matrix with normalized data data_mean: vector of mean used to normalize t

(normalizedData, data_mean, data_std, dimensions_to_ignore, actions, one_hot)

def unNormalizeData(normalizedData, data_mean, data_std, dimensions_to_ignore, actions, one_hot):
    """Borrowed from SRNN code. Reads a csv file and returns a float32 matrix.
    https://github.com/asheshjain399/RNNexp/blob/srnn/structural_rnn/CRFProblems/H3.6m/generateMotionData.py#L12

    Args
      normalizedData: nxd matrix with normalized data
      data_mean: vector of mean used to normalize the data
      data_std: vector of standard deviation used to normalize the data
      dimensions_to_ignore: vector with dimensions not used by the model
      actions: list of strings with the encoded actions
      one_hot: whether the data comes with one-hot encoding
    Returns
      origData: data originally used to
    """
    T = normalizedData.shape[0]
    D = data_mean.shape[0]

    origData = np.zeros((T, D), dtype=np.float32)
    dimensions_to_use = []
    for i in range(D):
        if i in dimensions_to_ignore:
            continue
        dimensions_to_use.append(i)
    dimensions_to_use = np.array(dimensions_to_use)

    if one_hot:
        origData[:, dimensions_to_use] = normalizedData[:, :-len(actions)]
    else:
        origData[:, dimensions_to_use] = normalizedData

    # potentially ineficient, but only done once per experiment
    stdMat = data_std.reshape((1, D))
    stdMat = np.repeat(stdMat, T, axis=0)
    meanMat = data_mean.reshape((1, D))
    meanMat = np.repeat(meanMat, T, axis=0)
    origData = np.multiply(origData, stdMat) + meanMat
    return origData


def revert_output_format(poses, data_mean, data_std, dim_to_ignore, actions, one_hot):