What you will learn in Data Analysis with LLMs
Text, tables, images and sound
Chapter by chapter
Concepts explained in this book
- language modelA language model with a very large number of learned parameters, trained on broad text data for general-purpose language tasks.
- classifierClassification is the assignment of one or more predefined categories to an input item.
- SQLSQL is a standardized language for defining, querying, and manipulating data in relational database systems.
- GPT-4oGPT-4o is OpenAI's multimodal generative model in the GPT family.
- ChatGPTChatGPT is a conversational artificial-intelligence service that lets users interact with OpenAI language models through natural-language messages.
- promptA textual instruction supplied to a language model that specifies a task, context, inputs, and often the desired output format.
- toolA tool is a callable software function whose interface is described in natural language so that a language model can invoke it.
- templateA prompt template is a reusable text specification containing placeholders that can be replaced with input-dependent content to produce a prompt.
- AI agentAn agent is a software pattern in which a language model dynamically decomposes a task, selects actions or tools, and plans subsequent steps from their results.
- text-to-SQLText-to-SQL translation is the task of translating questions expressed in natural language into SQL queries.
- relational database management systemA relational database management system is software that stores, manages, and processes relational data organized in tables.
- SQLiteSQLite is a lightweight relational database engine that stores an entire database in a self-contained file and supports SQL.
- database tableRelational data is data organized as collections of tables whose rows and columns represent related records and attributes.
- OpenAI API SDKAn OpenAI client is an object that represents a configured programmatic connection to OpenAI model services.
- fine-tuningFine-tuning is the process of further training an existing model on task-specific examples to adapt its behavior to a particular task.
- LangChainLangChain is an open-source software framework for developing applications that integrate language models with prompts, data, and external actions.
- databaseA structured collection of data organized for storage, retrieval, and management by computer software.
- summarizationText summarization is the production of a shorter text that preserves the most important information from a source text.
- placeholderA missing-value placeholder is a marker stored in place of a value that is unavailable in an extracted record.
- chat completionA chat completion is an API operation that generates an assistant response from a sequence of conversational messages.
- output formatStructured output is a result organized according to an explicit schema or format so that its components can be processed programmatically.
- parameterA model parameter is a learned numerical value that determines a machine-learning model's behavior.
- schemaA database schema is the defined structure of a database, including its tables, columns, data types, and relationships.
- structured dataData organized according to a predefined schema, such as tables with named rows and columns or graphs with entities and relationships.
- fine-tuned modelA fine-tuned model is a model obtained by adapting an existing model through additional training on task-specific data.
- embeddingAn embedding is a numerical vector representation of an object that is intended to preserve meaningful properties of that object.
- dialogueA message history is an ordered collection of conversational messages representing an interaction with a chat model.
- pipelineA data-processing pipeline is an ordered sequence of computational stages that transform or analyze data, with each stage passing results to the next.
- custom modelA language model trained or adapted for particular tasks or domains.
- temperatureTemperature is a model-generation setting that controls the degree of stochastic variation in selecting output tokens from their probability distribution.
- MapA Python dictionary is a mutable mapping data structure that associates keys with values.
- LlamaIndexLlamaIndex is an open-source software framework for building applications that connect language models to external data and retrieval systems.
- structured extractionInformation extraction is the task of identifying and structuring specified facts or entities from text.
- sounddevicesounddevice is a Python library for recording and playing audio through sound cards and other audio devices.
- digital imageImage data is data represented as visual images containing depicted objects, people, scenes, or other visual information.
- prompt engineeringPrompt engineering is the systematic design and search of prompts or prompt templates to guide a language model toward better task results.
- data analysisData analysis is the process of examining, transforming, and interpreting data to discover patterns, answer questions, and derive insights.
- speech recognitionSpeech recognition is the automatic conversion of spoken language into a machine-readable representation, commonly text.
- few-shot learningFew-shot learning is an approach in which a model infers a task from a small number of examples specifying inputs and desired outputs.
- training datasetTask-specific training data is a dataset of examples tailored to the task for which a model is being adapted.
- attributeA graph property is a named attribute associated with a graph node or relationship and holding a value.
- Base64Base64 encoding is a binary-to-text encoding scheme that represents binary data using a restricted set of printable characters.
- graphA graph is a data structure that represents entities as nodes and relationships as edges between them.
- machine translationTranslation is the conversion of content from a source language into an equivalent target language.
- database result setA query result is the table of rows and columns produced by executing a database query.
- voice query interfaceA software interface that accepts spoken questions about structured data and returns answers by converting speech into executable data queries.
- Whisperwhisper-1 is the model identifier for the Whisper speech-transcription model in the OpenAI API.
- tokenA token is the atomic unit of text representation processed by a language model, typically corresponding to about four characters.
- web searchWeb search is the process of finding information on the World Wide Web in response to a query.
- natural language interfaceA natural language query interface converts questions expressed in human language into database queries and returns answers from the queried data.
- post-processingPostprocessing is the processing of generated outputs after model generation to filter, evaluate, or select results.
- SciPySciPy is a Python library providing scientific-computing algorithms and data structures, including functionality for processing numerical and signal data.
- visual question answeringA visual question-answering system is a system that answers natural-language questions about the content of an image or other visual input.
- CypherA Cypher query is a declarative statement written in Cypher that requests or manipulates graph data.
- clusteringText clustering is the unsupervised grouping of text documents according to the similarity of their represented features.
- GPT-4GPT-4 is a GPT language model version capable of generating and understanding text, including complex code.
- hyperparameter tuningModel tuning is the adjustment of a model's generation settings to improve its behavior on a task.
- tokenizerA tokenizer is a software component that transforms text into the token identifiers used by a language model.
- data aggregationAn aggregate function computes a summary value from multiple rows or matched records in a query result.
- chat.completions.create()client.chat.completions.create is an OpenAI Python library method that submits a chat completion request.
- ClaudeClaude 3.5 is a version of Anthropic's Claude family of large language models.
- text-to-speech conversionSpeech generation is the production of spoken audio from textual or other input content.
- transformersHugging Face Transformers is an open-source library providing pretrained Transformer models and tools for natural-language, vision, and multimodal machine-learning tasks.
- audioAn audio file is a digital file that stores recorded sound for playback or processing.
- UTF-8UTF-8 is a variable-length character encoding standard that represents Unicode text using one to four bytes per character and preserves ASCII encoding for the first 128 characters.
- aria semanticsA role attribute identifies the source type of a chat message, such as user, assistant, or system.
- regular expressionA regular expression is a pattern notation for matching and extracting substrings from text.
- randomizationRandomization is the introduction of stochastic variation during generation so repeated generation attempts can produce different outputs.
- target languageA target language is the language or language variant into which source content is translated.
- labelA class label is a category designation assigned to an instance in a classification task.
- videoVideo data consists of sequences of visual frames, typically accompanied by motion and possibly audio.
- voice inputSpoken input is information conveyed to a computer through human speech rather than typed text.
- task descriptionA textual specification of the operation that a model should perform.
- transformer architectureThe Transformer is a neural-network architecture designed to model relationships in sequences, especially sequences with variable-length inputs or outputs.
- nodeA node is an entity in a graph that can be connected to other nodes and associated with properties.
- graph databaseA graph data management system is software that stores and processes graph data consisting of entities and relationships.
- model selectionModel selection is the process of choosing a model based on task requirements, performance, cost, data sensitivity, and deployment constraints.
- CSVCSV is a text-based tabular-data format in which each row occupies a line and column values are separated by commas.
- multimodal dataMultimodal data combines information from multiple modalities, such as text, images, audio, or video.
- tool decoratorThe LangChain tool decorator is a Python decorator that converts a function into a tool with metadata describing its name, parameters, and behavior.
- GeminiGemini is Google's family of multimodal generative AI models for processing and generating content.
- SQLite command-line interfaceThe SQLite command-line interface is an interactive program for creating, querying, and managing SQLite databases through text commands.
- connectorAn integration component that enables a language model to access external data sources such as web search services or databases.
- Command R+Command R+ is a large language model developed by Cohere for generation, retrieval-augmented generation, and related language tasks.
- playsoundplaysound is a Python library for playing audio files through the system's audio output.
- formal languageA formal language is a language defined by explicit symbols and grammatical rules for expressing structured statements to computational systems.
- top-p filteringA top_p parameter is a model-generation setting that limits eligible output tokens to a cumulative probability mass.
These guides are generated from the text of Manning books by Manning's knowledge graph and link to the sections they draw on. The full books are available on liveBook and at manning.com.