charlesbmi

from typing import Annotated, Union

import numpy as np
from pydantic import (
    AfterValidator,
    BaseModel,
)

FloatInfAsNone = Annotated[
    Union[float, None], AfterValidator(lambda x: None if x == np.inf else x)

charlesbmi

#!/bin/bash
#
# General Usage:
#     ./rsync_script.sh [additional rsync options] <source> <destination>
# Example with Additional Options:
#     ./rsync_script.sh --bwlimit=1000 --info=progress2 /path/to/source user@destination:/path/to/destination


# Check if sufficient arguments are provided
if [ $# -lt 2 ]; then

charlesbmi

# 1. Reconstruction loss: x0 = x0_recon
reconstruction_loss = tf.keras.losses.mean_squared_error(x0, x0_reconstructed)
model.add_loss(reconstruction_loss)

# 2. Future state prediction loss: x1 = x1_pred
state_prediction_loss = tf.keras.losses.mean_squared_error(x1, x1_pred)
model.add_loss(state_prediction_loss)

# 3. Linear dynamics loss: y1 = K * y0
linear_dynamics_loss = tf.keras.losses.mean_squared_error(code1, code1_pred)

charlesbmi

# x_train, y_train = ...

model.compile(loss=tf.keras.losses.CategoricalCrossentropy())
model.fit(x=x_train, y=y_train)

charlesbmi

# Add onto our existing x0/code0 computation graph
# x1 is the input x_{k+1}
x1 = tf.keras.Input(shape=INPUT_DIM, name='x1')
# we can re-use the same sub-models (encoder, decoder, linear_dynamics)
# to share the same weights
code1 = encoder(x1)
x1_pred = decoder(code1_pred)

model = tf.keras.Model(
    inputs={'x0': x0, 'x1': x1},

charlesbmi

# Connect sub-models
x = tf.keras.Input(shape=INPUT_DIM)
code = encoder(x)
x_reconstructed = decoder(code)
next_code_pred = linear_dynamics(code)

model = tf.keras.Model(
    inputs=x,
    outputs={'x0_reconstructed': x_reconstructed,
             'y1': next_code_pred}

charlesbmi

import tensorflow as tf

# Hyperparameters
INPUT_DIM = 2  # Same as decoder output dimension
HIDDEN_DIM = 30
LATENT_DIM = 2

# Encodes input to low-dimensional code
encoder = tf.keras.Sequential(
    [

charlesbmi

charlesbmi

pr <- function(...) {
  # prints variable names and values
  dots <- substitute(list(...))[-1]  # remove 1st element, "list"
  print(sapply(dots, deparse))  # names
  print(paste(list(...)))  # values
}

# demo
a <- 1
b <- c(3,5,2)