geyang

    1  df
    2  ls /dev/sdh
    3  cd /dev/sdh
    4  sudo mount /dev/sdb1 ~/runs
    5  mkdir runs
    6  sudo mount /dev/sdb1 ~/runs
    7  sudo mount /dev/sh~/runs
    8  sudo mount /dev/sdh ~/runs
    9  ls
   10  cd runs

geyang

Example for Rendering the Landmarks

geyang

ffmpeg -i input.mp4 -filter:v "crop=674:64:0:0" output.mp4

geyang

diff --git a/mac/mac_vanilla.py b/mac/mac_vanilla.py
index 1565db7..28d5970 100644
--- a/mac/mac_vanilla.py
+++ b/mac/mac_vanilla.py
@@ -1,3 +1,5 @@
+from copy import deepcopy
+
 import torch
 from torch import nn
 from torch import optim

geyang

The new 1.10 version of hdf5 library is stable. You can now install from the source to use this new capability.

The installation process has a few caveats, so here is how to do it.

Note:

1. you need to first remove the hdf5 libraries currently installed in ubuntu.
2. Then you need to reinstall h5py.
3. when you reinstall h5py via pip, use the --no-binary flag. This is because the h5py wheel comes with its own hdf5 binary.

Instructions:

geyang

class MultiHeadedMlp(nn.Module):
    def __init__(self, input_dim, *out_dims):
        super().__init__()
        self.out_dims = out_dims
        self.heads = nn.ModuleList([nn.Linear(input_dim, dim) for dim in out_dims])

    def forward(self, i):
        return [h(i) for h in self.heads]

geyang

#!/bin/bash

echo "Restarting bluetooth service..."
blueutil -p 0 && sleep 1 && blueutil -p 1

echo "Waiting bluetooth service to be restored..."
until blueutil -p | grep "1" >/dev/null; do sleep 1; done

echo "Searching for devices not connected..."
devices=($(blueutil --paired | grep "not connected" | awk -F '[ ,]' '{print $2}'))

geyang

firs create a group, add users to that group se the folder to this group, recursively if this takes too long, use glob patterns to select the top k-levels

1. sudo chgrp -R /checkpoint/geyang
2. sudo chmod -R 2777 /checkpoint/geyang
3. when running logger, use umask 0 to make sure all files are written with 777 permission.

For all instances writing, really need to umask 0 to make sure that files are writable from multible accounds. For exisiting files written by different user accounts, use chmod -R 2777 to amend the permissions.

geyang

version: 0
hosts:
  fair: &devfair_host
    ip: localhost
    port: 41000
    username: geyang
    pem: ~/.ssh/fair-loaner
mounts:
  - !mounts.SSHCode &plan2vec_code
    <<: *devfair_host

geyang

List of Data Loader Patterns for Tensorflow

These code come from the Tensorflow documentation the (experimental) new Dataset API.

General types:

• one_shot
• initializable
• reinitializable
• feedable