Datasets:
license: mit
language:
- en
tags:
- Dog
- Diseases
- Symptoms
- Pet
- Canine
- RAG
- Care
- Veterinary
- Breeding
- Pet-health
- Dog-health
- Health
size_categories:
- 1K<n<10K
task_categories:
- text-generation
- question-answering
- summarization
Dataset Card for Everything About Dogs Dataset
This dataset is built from the book Everything About Dogs
Everything about dogs is a book by AL G Eberhart.
The book contains topics on diseases, how to feed, environment to provide, how to take care of pups and everything about dogs.
Motivation
In Memory of Caspu
Dogs can't speak but they can express their pain. However this expression is often misunderstood or ignored.My I aim is to build a retrieval engine from trusted resources which can provide answers to me everything about dogs.
Dataset Details
Dataset Description
There are two datasets here.
Dataset Sources
- Repository: Vet RAG Dataset Github - theraykar
- Book: Everything About Dogs
Uses
Direct Use
The disease dataset can be used to identify diseases based on symptoms using RAG.
While the everything_about_dogs dataset can be used to get answers related to taking care of dogs.
Out-of-Scope Use
This dataset contains information from a 1902 book and curating with a llm search. The information may be outdated.
Dataset Structure
There are two datasets the disease dataset and book dataset. Following is the structure
The book dataset is made of primary two files. The index json and the everything_about_dogs_with_embedding.parquet file.
The index json contains chapter names which are files as keys. These files are stored in chapter folder.
The values are summary. Following is sample data
{
"everything_about_dogs_administering_medicine.txt": "Here is a summary of the document everything_about_dogs_administering_medicine.txt:\n\n* It is better to trick a dog into taking medicine naturally rather than forcing it.\n* Liquids can be poured into the patient's cheek to drain through the teeth.\n* To make a dog swallow, hold their head up, close their nostrils and jaws for a couple of seconds.\n* Pills can be placed at the back of the tongue, and the jaws closed quickly to make the dog swallow.\n* Pills can be hidden in a piece of meat.\n* Castor oil, cod liver oil, syrup of phosphates, and many powders can be mixed with savory broth or porridge.\n* Powders can be mixed with butter, honey or syrup, and placed at the back of the tongue.\n* Gelatine capsules are useful for administering powders.\n* Use a rubber syringe, not a glass one, for injections.\n* Dogs resemble man in temperament and nervous organization, but their digestive organs differ, so medicine does not act on each alike.\n* The dog is particularly susceptible to the action of chloroform.\n* Kindness and strategy are the only safe methods for giving medicine.\n* Hypodermic injections of alkaloidal solutions are the most prompt, efficient and reliable method of administration.\n* For ordinary purposes, tablets or capsules are the most convenient method.\n* Liquid medicines should be avoided if possible, but if necessary, Rochelle salts are the least nauseating saline laxative.\n* Castor oil can be given in about four parts of milk to disguise the taste.\n* For nasal passages, use warm water and steaming.\n* For the ear, use a blunt, hard rubber syringe or a \"blower\" for dry powder.\n* In treating wounds, antisepsis is important, but avoid iodoform.\n* In the application of unguents to the skin of long-haired dogs, clipping may be necessary.\n* Rectal injections are indicated in cases of collapse, extreme weakness and refusal of food.\n* All drugs act more powerfully on an empty stomach.\n* Powerful remedies and those which are of irritating character should be given after feeding.",
"everything_about_dogs_breeding.txt": "Here is a summary of the document everything_about_dogs_breeding.txt in points:\n\n* Bitches typically come into season twice a year, and breeding during the first oestrum depends on breed and maturity. It's generally advised to wait until at least eighteen months for larger breeds.\n* Breeding a bitch at every period is not recommended unless she is exceptionally vigorous; allowing service every other period is preferable.\n* Signs of a bitch being in season include bleeding, with readiness for the dog typically between the tenth and twelfth day from the start of bleeding.\n* Gestation lasts approximately 63 or 64 days, with variations possible, but veterinary assistance may be needed for significant deviations.\n* Spring is considered the best time for mating, allowing puppies to grow strong during the summer.\n* Regulated exercise is crucial both at mating and throughout pregnancy, adjusted to the individual bitch's needs.\n* A bitch should be in good condition before oestrum to produce a healthy litter, with attention to diet and health three months prior if needed.\n* A mild cathartic may be beneficial before whelping to ensure proper bowel condition.\n* One service is usually sufficient when the bitch is ready, but a second service after 24 hours can provide assurance.\n* After mating, the bitch needs careful attention, including seclusion and regulated exercise.\n* Diet should be adjusted as the bitch becomes heavy with pups, increasing the amount and providing beef, mutton, bread, milk and soups.\n* Loss of appetite after giving birth can be addressed with assafetida to restore appetite and milk flow.\n* Provide a secluded and prepared stall or kennel a week before the due date, ensuring warmth, light, and ventilation.\n* Secure an old carpet or blanket to the floor of the stall to provide puppies with a foothold.\n* Protect pups from being laid on by fastening a shelf a few inches from the floor to the sides and back of the box.\n* Minimize disturbance during whelping, but be prepared to provide assistance or veterinary care if complications arise.\n* Avoid unnecessary force if manual assistance is required, and consider Fellows' Compound Syrup of Hypophosphate to aid labor pains if needed.\n* Dangers are greater when a bitch is bred to a much larger dog or in finely bred, pampered bitches.\n* Consult a veterinary surgeon for wrong presentations or disproportionate pup size.\n* Ensure adequate milk supply; if inadequate, use a foster mother or artificial food like Spratt's Orphan Puppy Food.\n* Weaning should commence at six to seven weeks, gradually separating the bitch and feeding pups goat's or cow's milk, boiled with water, and bread. Condensed milk is preferred.\n* After weaning, supplement with oatmeal, wheat middlings, cornmeal, and lime water for bone formation.\n* Address bone weakness with precipitated phosphate of lime for the mother.\n* Gastro-enteritis in puppies should be treated with castor or sweet oil and opiates.\n* Dr. A.J. Sewell's advice includes using parturition forceps or a crochet for difficult deliveries, avoiding hot baths, and considering an ice bag or a drive to stimulate labor.\n* Prolapsus or inversion of the womb requires veterinary attention, and amputation may be necessary in severe cases.\n* Post-whelping care involves providing fresh bedding, warm milk, and easily digestible foods like oatmeal gruel.\n* Ensure the bitch goes out to relieve herself to prevent complications.\n* Address issues like excessive discharge with astringent injections or mammary gland problems with hot water fomentations and camphorated oil.\n* Parturient eclampsia can be managed with bromide of potassium or hydrated chloral.\n* Maintain cleanliness in puppy quarters to minimize flies.\n* Protect the bench on all sides and keep it at the right size, preventing puppies from getting cold before drying.",
"everything_about_dogs_diseases.txt": "......",
}
This can guide the user or any agentic RAG entity to locate the source to lookup based on query.
Once the file name is determined, another file everything_about_dogs_with_embedding.parquet file helps to get the textual chunks or embeddings for the given file.
The paraquet file has columns chapter_name, text chunks and embedding list.
Following is a json preview (data not real) of another file everything_about_dogs_with_embedding.json
{
"everything_about_dogs_administering_medicine.txt":{
"text_chunks": ["....","..."],
"embedding" :[[1, 2,...],[1, 2.1, ...]]
},
"everything_about_dogs_breeding.txt":{
"text_chunks": ["....","..."],
"embedding" :[[2, 1.1,...],[0.2, 1.1,...]]
},
}
The disease dataset set is on diseases extracted from the Diseases chapter of the book.
The file diseases_dataset_with_embeddings.parquet or .json
It contains keys or columns as follows
raw_description: "The description and name of the diseases extracted from the book"
references: List of diseases referred by the given disease.
symptoms: Symptoms extracted from the raw data from book by llm. If not available the using search by llm.
when_to_approach_vet: Conditions which tell when to approach the vet.
food_during_disease: Food to provide during dog is ill.
embedding: Embeddings created from all the top keys and values using 'all-MiniLM-L6-v2'
Dataset Creation
Source Data
Data Collection and Processing
First the text is extracted from the book Everything About Dogs
The manually each chapter is seperated.
For each chapter the text is cleaned.
For the disease chapter diseases are recognised using regex and the disease to text mapping is established.
Any unknown disease referenced is recognised and its information is also added.
Finally LLM is called in every disease data to structure and extract symptoms, when to see vet, food.
The Large Language model used is Gemini Flash 2.0 and Pydantic AI is used for structuring.
Coming to embedding, all-MiniLM-L6-v2 is used to embed all diseases as well as chunks of chapter.
Each chunk for embedding chapter is of size 2000 charachters with overlap of 1000 charachters. The endings are ignored.
The embeddings can be ignored. And custom embeddings can be performed by joining the text chunks.
You have a look at the notebooks here Vet RAG Dataset Github - theraykar
Bias, Risks, and Limitations
The data comes from the book and therefore will have author bias.
Diseases referenced and some data not available are filled using duckducksearch LLM Agent.
The quality of data is therefore mixed. I will add a future tag of what is filled in.
To get the reference from where the chunk came from the disease dataset also has the raw text column. This can be used for validation.