noahtren

@functools.partial(nn.scan,
                   variable_broadcast='params',
                   split_rngs={'params': False})
@nn.remat
@nn.compact
def __call__(self, carry, t: int):
  noise, memory, latent_code = carry['noise'], carry['memory'], carry[
      'latent_code']
  intermediates = {}

noahtren

import matplotlib.pyplot as plt
import numpy as np


def normalize_freq_coeff(coeff):
  freqs = np.fft.fftfreq(coeff.shape[-1], d=1 / coeff.shape[-1])
  scale = (1 / (np.abs(freqs) + 1))
  return coeff * scale

noahtren

videoId = 'QRE0GxT6Zbw'
response = requests.get(f'http://video.google.com/timedtext',
  params={
    'lang': 'en',
    'v': videoId,
    'fmt': 'vtt'
  })
print(response.text)

noahtren

import tensorflow as tf
import tensorflow_text as text
from transformers import AutoTokenizer

def get_tf_tokenizer(hf_model_name, do_test=False):
  hf_tokenizer = AutoTokenizer.from_pretrained(hf_model_name)
  model_proto = hf_tokenizer.sp_model.serialized_model_proto()
  tf_tokenizer = text.SentencepieceTokenizer(model=model_proto, out_type=tf.int32)
  if do_test:
    test_string = "This is a testtt, hah! reaaly cool :)"

noahtren

Colab + S3 For Syncing Local Code

Scripts to add to your Colab Notebook to sync with an S3 bucket that contains code to run.

---

1. Write AWS credentials to disk in order to use s3fs

%%writefile ~/.passwd-s3fs
:

noahtren

"""Running this uploads all Python modules in this workspace to an S3 bucket, to
access when training on Colab.
"""

import os

import boto3

S3_BUCKET_NAME = 'BUCKET-NAME'
# files in the local directory to upload as code. (yaml or json are often used for config)

noahtren

"""Originally from https://www.kaggle.com/cdeotte/rotation-augmentation-gpu-tpu-0-96#Data-Augmentation,
modified to be in pure TensorFlow and to work on a batch of images rather than a single image.
(For a tf.data pipeline, you may want to look at the original code at the link above.)
"""

import math

import tensorflow as tf

def transform_batch(images,

noahtren

"""Client class to interface with FoodData Central.
FoodData Central requires a Data.gov key: https://api.data.gov/signup/
"""

from enum import Enum
import json

import requests

BASE_URL = 'https://api.nal.usda.gov/fdc/v1'

noahtren

Keybase proof

I hereby claim:

• I am noahtren on github.
• I am noahtren (https://keybase.io/noahtren) on keybase.
• I have a public key ASAiy9q-1SBwnj5ssQx5Zx3LOVQayhLJmWXSAGSAlp1Hjwo

To claim this, I am signing this object:

noahtren

"""Small convnet with residual connections.
inspired by https://gist.github.com/mjdietzx/0cb95922aac14d446a6530f87b3a04ce, 
which builds a full ResNet-50 or ResNeXt-50 model
"""

NUM_CLASSES = 2
from keras.layers import BatchNormalization, Conv2D, LeakyReLU, Input, MaxPool2D, Dense, Flatten, Dropout
from keras.models import Model

def add_common_layers(y):