great-ai/examples/simple/data.ipynb

231 lines
6.8 KiB
Text

{
"cells": [
{
"cell_type": "markdown",
"metadata": {},
"source": [
"# Train a domain classifier on the [semantic scholar dataset](https://api.semanticscholar.org/corpus)\n",
"> Part 1: obtain and clean data\n",
"\n",
"![position of this step in the lifecycle](diagrams/scope-data.svg)\n",
"> The blue boxes show the steps implemented in this notebook."
]
},
{
"cell_type": "code",
"execution_count": 1,
"metadata": {},
"outputs": [],
"source": [
"MAX_CHUNK_COUNT = 4"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"### Extract\n",
"\n",
"This can be achieved by downloading a public dataset (such as in this case), or by having a Data Engineer setup and give us access to the organisation's data.\n",
"\n",
"In this case, we download the semantic scholar dataset from a public S3 bucket."
]
},
{
"cell_type": "code",
"execution_count": 2,
"metadata": {},
"outputs": [
{
"data": {
"text/plain": [
"'Processing 4 out of the 6002 available chunks'"
]
},
"execution_count": 2,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"import urllib.request\n",
"\n",
"manifest = (\n",
" urllib.request.urlopen(\n",
" \"https://s3-us-west-2.amazonaws.com/ai2-s2-research-public/open-corpus/2022-02-01/manifest.txt\"\n",
" )\n",
" .read()\n",
" .decode()\n",
") # a list of available chunks separated by '\\n' characters\n",
"\n",
"chunks = manifest.split()[:MAX_CHUNK_COUNT]\n",
"\n",
"f\"Processing {len(chunks)} out of the {len(manifest.split())} available chunks\""
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"### Transform\n",
"\n",
"- Filter out non-English abstracts using `great_ai.utilities.predict_language`\n",
"- Project it to only keep the necessary components (text and labels), clean the textual content using `great_ai.utilities.clean`\n",
"- We will speed up processing using `great_ai.utilities.parallel_map`."
]
},
{
"cell_type": "code",
"execution_count": 3,
"metadata": {},
"outputs": [
{
"name": "stderr",
"output_type": "stream",
"text": [
"\u001b[38;5;226m2022-06-19 14:59:12,562 | WARNING | Limiting concurrency to 4 because there are only 4 chunks\u001b[0m\n",
"\u001b[38;5;39m2022-06-19 14:59:12,563 | INFO | Starting parallel map (concurrency: 4, chunk size: 1)\u001b[0m\n"
]
},
{
"data": {
"application/vnd.jupyter.widget-view+json": {
"model_id": "ff8fc113515944cfa75127f4aba3d491",
"version_major": 2,
"version_minor": 0
},
"text/plain": [
" 0%| | 0/4 [00:00<?, ?it/s]"
]
},
"metadata": {},
"output_type": "display_data"
}
],
"source": [
"from typing import List, Tuple\n",
"import json\n",
"import gzip\n",
"from great_ai import parallel_map, clean, is_english, predict_language\n",
"\n",
"\n",
"def preprocess_chunk(chunk_key: str) -> List[Tuple[str, List[str]]]:\n",
" # Extract\n",
" response = urllib.request.urlopen(\n",
" f\"https://s3-us-west-2.amazonaws.com/ai2-s2-research-public/open-corpus/2022-02-01/{chunk_key}\"\n",
" ) # a gzipped JSON Lines file\n",
"\n",
" decompressed = gzip.decompress(response.read())\n",
" decoded = decompressed.decode()\n",
" chunk = [json.loads(line) for line in decoded.split(\"\\n\") if line]\n",
"\n",
" # Transform\n",
" return [\n",
" # Create pairs of `(text, [...domains])`\n",
" # The text is cleaned to remove PDF extraction, web scraping, and other common artifacts\n",
" (\n",
" clean(\n",
" f'{c[\"title\"]} {c[\"paperAbstract\"]} {c[\"journalName\"]} {c[\"venue\"]}',\n",
" convert_to_ascii=True,\n",
" ),\n",
" c[\"fieldsOfStudy\"],\n",
" )\n",
" for c in chunk\n",
" if c[\"fieldsOfStudy\"] and is_english(predict_language(c[\"paperAbstract\"]))\n",
" ]\n",
"\n",
"\n",
"preprocessed_chunks = parallel_map(preprocess_chunk, chunks)"
]
},
{
"cell_type": "code",
"execution_count": 4,
"metadata": {},
"outputs": [],
"source": [
"from itertools import chain\n",
"\n",
"preprocessed_data = list(chain(*preprocessed_chunks))\n",
"X, y = zip(\n",
" *preprocessed_data\n",
") # X is the input, y is the expected (ground truth) output"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"### Load\n",
"\n",
"Upload the dataset (or a part of it) to a central repository using `great_ai.add_ground_truth`. This step automatically tags each datapoint with a split label according to the ratios we set. Additional tags can be also given.\n",
"\n",
"#### Use a different repository\n",
"\n",
"For the sake of simplicity, the tutorial uses the local hard drive (`great_ai.ParallelTinyDbDriver`) as the central repository.\n",
"This can be simply changed, for example, by the following snippet:\n",
"\n",
"```python\n",
"from great_ai import configure, MongoDbDriver\n",
"\n",
"configure(tracing_database=MongoDbDriver('mongodb://localhost:27017_or_something_like_that'))\n",
"```"
]
},
{
"cell_type": "code",
"execution_count": 5,
"metadata": {},
"outputs": [
{
"name": "stderr",
"output_type": "stream",
"text": [
"\u001b[38;5;226m2022-06-19 15:03:30,300 | WARNING | Environment variable ENVIRONMENT is not set, defaulting to development mode ‼️\u001b[0m\n",
"\u001b[38;5;226m2022-06-19 15:03:30,301 | WARNING | The selected persistence driver (ParallelTinyDbDriver) is not recommended for production\u001b[0m\n",
"\u001b[38;5;39m2022-06-19 15:03:30,301 | INFO | Options: configured ✅\u001b[0m\n"
]
}
],
"source": [
"from great_ai import add_ground_truth\n",
"\n",
"add_ground_truth(X, y, train_split_ratio=0.8, test_split_ratio=0.2)"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"### Next: [Part 2](train.ipynb)"
]
}
],
"metadata": {
"kernelspec": {
"display_name": "Python 3.10.4 ('.env': venv)",
"language": "python",
"name": "python3"
},
"language_info": {
"codemirror_mode": {
"name": "ipython",
"version": 3
},
"file_extension": ".py",
"mimetype": "text/x-python",
"name": "python",
"nbconvert_exporter": "python",
"pygments_lexer": "ipython3",
"version": "3.10.4"
},
"orig_nbformat": 4,
"vscode": {
"interpreter": {
"hash": "02dd6d3afbfa9fbbe1037d64ad9014965528a1ccad21929d6e72f466389a68ad"
}
}
},
"nbformat": 4,
"nbformat_minor": 2
}