Machine Learning on Google Cloud: Sequence and Text Models

Begin with a Gmail account and log into Google Cloud Platform. Learn about free credits, pricing tools, and core services like Compute Engine, BigQuery, Cloud SQL, Dataflow, Kubernetes, and AutoML.

Explore the GCP interface and learn how to start a free trial with a $300 credit. Navigate services like Compute Engine, Kubernetes, BigQuery, Cloud Storage, Cloud SQL, and SQL workspace.

Configure access for Google Cloud projects using the Cloud Resource Manager, by managing IAM permissions, assigning roles (owner, editor, viewer), and controlling principals.

Explore the core components of the GCP machine learning ecosystem, from data ingestion with Cloud Pub/Sub to processing with Dataflow and BigQuery, plus ML tools like TensorFlow and AutoML.

Discover Google Cloud's machine learning ecosystem, from Cloud Pub Sub ingestion to data processing with Cloud Dataflow, BigQuery, and Cloud Storage, plus TensorFlow, AI platform, Dialogflow, TensorFlow Enterprise, AutoML.

Set up a Google Cloud bucket, configure region and storage class, ensure public access is prevented, and upload data to enable access from BigQuery and SQL workspace in ML code.

Learn how to set up a Google Cloud Compute Engine virtual machine, choose machine types and disks, and explore templates, images, and marketplace options for ML workloads.

Explore the nuts and bolts of Google BigQuery, manage datasets, run SQL queries on big data, and work with Covid and Italy datasets, open source marketplace data, and transfers.

Use Jupyter notebooks on Vertex AI to prepare data, access cloud data sources, and build text and sequence models with pre-trained options like palm and embeddings.

Learn to manage Google Cloud Jupyter Lab notebook instances, create notebooks with region and machine type options, and monitor GPUs, costs, and Python environments like numpy, PyTorch, TensorFlow.

Access Jupyter Lab quickly from the console by viewing notebook data, right-clicking to open a new tab, and launching your instances. Then use Jupyter Lab and GCP for textual analysis.

Prepare your environment by pre-installing TensorFlow for machine learning on Google Cloud to support sequence and text models.

Access data from a Google Cloud storage bucket in Jupyter Lab via managed notebooks, initialize a storage client, and list pdf reports stored under the reports folder.

Learn to run Jupyter notebooks in Google Colab, manage Colab Notebooks in Google Drive, create and run code cells, and load data from GitHub or uploads into pandas dataframes.

Explore how Google Colab provides access to GPU and TPU resources for training neural networks, how to enable a hardware accelerator, and how GPU vs CPU affects deep learning workloads.

Access a single CSV from a GCP bucket into Colab by authenticating, installing Google Cloud Storage, creating a storage client, downloading the CSV, and loading it with pandas.

Read and process multiple PDFs from a Google Cloud Storage bucket by authenticating in Colab, extracting text from each PDF, and compiling pdf_texts for analysis.

Access the OpenAI API for text models and large language models by creating and managing API keys, logging in, and configuring billing, usage, and payment methods.

Sign up for Hugging Face, log in, and explore the Falcon 1 tb demo while creating access tokens to authenticate to the Hugging Face hub API for large language models.

Explore LangChain, a Python library that enables building NLP-to-LM applications by chaining prompts, memory, and agents. Connect to OpenAI GPT-4 and other APIs and use text splitters and vector stores.

Leverage Google Cloud to build sequence and text models and learn to read and process PDF documents for machine learning workflows.

Learn how to read in multiple PDFs for sequence and text models on Google Cloud.

Learn to clean text data by removing punctuation and English stopwords, lowercase and tokenize, and apply tf-idf and idf vectorization to prepare chunks from PDF text.

Tokenize the text with the nltk-like package, convert tokens to lowercase, and remove punctuation and stopwords to create clean data for analysis.

Set up and manage your OpenAI API access to run text models and large language models, including creating and safeguarding API keys, reviewing usage and billing, and configuring payment methods.

Extract keywords from pre-processed text chunks using tf-idf with a tf-idf vectorizer, compute scores, and identify the top keywords for each chunk.

Identify relevant keywords across documents using tf-idf to transform text into a numerical representation and highlight terms by term frequency and inverse document frequency, with stopword removal as pre-processing.

Compute tf-idf vectors for the pdf texts and apply cosine similarity to reveal document similarity after stopwords removal and text cleaning.

Identify textual similarity between two text chunks using tf-idf vectors, cosine similarity, and euclidean distances. Clean documents by removing stop words and special characters, then vectorize and compute pairwise similarities.

Apply transformers to evaluate textual similarity using sentence transformers and BERT base, computing cosine similarity and euclidean distances on document embeddings to compare cleaned documents.

Perform entity analysis on textual data to extract nouns and key phrases, then identify and print named entities using NLTK's tokenization, POS tagging, and chunking.

Explore named entity linking by connecting text entities to a knowledge base, as in linking Washington and United States to Wikipedia, using tokenization, stopword removal, and sentence segmentation.

Learn the basics of long short-term memory networks (LSTMs), a special kind of recurrent neural network with forget, update, and output gates, and apply them to cryptocurrency time-series data.

Classify news categories from headlines and descriptions using a preprocessing pipeline that loads data from Google Cloud storage, builds a dataframe, and cleans text with stopwords removal and lemmatization.

Learn basic text data pre-processing for classification using one hot encoding and pad sequencing, including encoding the label variable with a label encoder, and splitting data with train_test_split.

Explore one hot encoding for categorical factors and the padding of text sequences to equal length, using word indices and a corpus for model input.

Tokenize training data with a 10,000-word vocabulary, convert text to integer sequences, and pad to length 20; build an embedding and an lstm with 64 units for a 35-class classifier.

Install transformers in a jupyter lab notebook, resolve dependencies, and load a pretrained DistilBERT uncased tokenizer in a PyTorch notebook for text classification.

Define a training loop in PyTorch that loads data in batches, runs forward and backward passes, and updates weights to minimize cross entropy loss with DistilBERT sequence classification across epochs.

Learn NumPy, the numerical Python library for multi-dimensional arrays and matrices, and use ndarray objects with import numpy as np for functions like np.array and np.zeros.

Explore how Pandas structures data with series and data frames, enabling data wrangling and analytics, and learn to read data from external sources for practical data science.

Import pandas, read a csv, and drop irrelevant columns. Impute missing ages by gender means and fill fare by its mean, while setting cabin to x and embarked to s.

Explore Python dictionaries as key-value data structures that store labeled data, demonstrating creation with curly brackets, access via keys, and operations like copy, delete, add, and sort.