Guest
Fundamentals in TensorFlow\ 5:21 Using TensorFlow Hub for pretrained word embeddings (transfer learning for NLP) 19:06 Mdl 6 Building, training and evaluating a transfer learning Mdl for NLP 29:52 Preparing subsets of data for Mdl 7 (same as Mdl 6 but 10% of data) 40:44 Mdl 7 Building, training and evaluating a transfer learning Mdl on 10% data 50:48 Fixing our data leakage issue with Mdl 7 and retraining it 1:04:31 Comparing all our modelling experiments evaluation metrics 1:17:46 Uploading our model’s training logs to TensorBoard and comparing them 1:29:01 Saving and loading in a trained NLP Mdl with TensorFlow 1:39:26 Downloading a pretrained Mdl and preparing data to investigate predictions 1:52:51 Visualising our model’s most wrong predictions 2:01:20 Making and visualising predictions on the test dataset 2:09:48 Understanding the concept of the speedscore tradeoff Project 2 SkimLit\ 2:24:50 Introduction to Milestone Project 2 SkimLit 2:39:10 What we’re going to cover in Milestone Project 2 (NLP for medical abstracts) 2:46:32 SkimLit inputs and outputs 2:57:34 Setting up our notebook for Milestone Project 2 (getting the data) 3:12:33 Visualising examples from the dataset (becoming one with the data) 3:25:51 Writing a preprocessing function to structure our data for modelling 3:45:42 Performing visual data analysis on our preprocessed text 3:53:37 Turning our target labels into numbers (ML models require numbers) 4:06:53 Mdl 0 Creating, fitting and evaluating a baseline Mdl for SkimLit 4:16:19 Preparing our data for deep sequence models 4:26:14 Creating a text vectoriser to map our tokens (text) to numbers 4:40:22 Creating a custom token embedding layer with TensorFlow 4:49:36 Creating fast loading dataset with the TensorFlow API 4:59:26 Mdl 1 Building, fitting and evaluating a Conv1D with token embeddings 5:16:48 Preparing a pretrained embedding layer from TensorFlow Hub for Mdl 2 5:27:41 Mdl 2 Building, fitting and evaluating a Conv1D Mdl with token embeddings 5:39:12 Creating a character-level tokeniser with TensorFlow’s TextVectorization layer 6:02:37 Creating a character-level embedding layer with 6:10:21 Mdl 3 Building, fitting and evaluating a Conv1D Mdl on character embeddings 6:24:07 Discussing how we’re going to build Mdl 4 (character token embeddings) 6:30:12 Mdl 4 Building a multi-input Mdl (hybrid token character embeddings) 6:45:49 Mdl 4 Plotting and visually exploring different data inputs 6:53:21 Crafting multi-input fast loading datasets for Mdl 4 7:02:02 Mdl 4 Building, fitting and evaluating a hybrid embedding model 7:15:20 Mdl 5 Adding positional embeddings via feature engineering (overview) 7:22:39 Encoding the line number feature to used with Mdl 5 7:35:05 Encoding the total lines feature to be used with Mdl 5 7:43:01 Mdl 5 Building the foundations of a tribrid embedding model 7:52:20 Mdl 5 Completing the build of a tribrid embedding Mdl for sequences 8:06:29 Visually inspecting the architecture of our tribrid embedding model 8:16:54 Creating multi-level data input pipelines for Mdl 5 with the API 8:25:54 Bringing SkimLit to life!!! (fitting and evaluating Mdl 5) 8:36:30 Comparing the performance of all of our modelling experiments 8:46:06 Saving, loading & testing our best performing model 8:53:55 Congratulations and your challenge before heading to the next module Series fundamentals in TensorFlow Milestone Project 3 BitPredict\ 9:06:29 Introduction to Milestone Project 3 (BitPredict) & where you can get help 9:10:23 What is a time series problem and example forecasting problems at Uber 9:18:09 Example forecasting problems in daily life 9:23:02 What can be forecast 9:31:00 What we’re going to cover (broadly) 9:33:35 Time series forecasting inputs and outputs 9:42:31 Downloading and inspecting our Bitcoin historical dataset 9:57:29 Different kinds of time series patterns & different amounts of feature variables 10:05:09 Visualizing our Bitcoin historical data with pandas 10:10:01 Reading in our Bitcoin data with Python’s CSV module 10:21:00 Creating train and test splits for time series (the wrong way) 10:29:38 Creating train and test splits for time series (the right way) 10:36:50 Creating a plotting function to visualize our time series data 10:44:47 Discussing the various modelling experiments were going to be running 10:53:59 Mdl 0 Making and visualizing a naive forecast model 11:06:15 Discussing some of the most common time series evaluation metrics 11:17:27 Implementing MASE with TensorFlow 11:27:05 Creating a function to evaluate our model’s forecasts with various metrics 11:37:16 Discussing other non-TensorFlow kinds of time series forecasting models 11:42:23 Formatting data Part 2 Creating a function to label our windowed time series 11:55:25 Discussing the use of windows and horizons in time series data
