rat-sql-gap/notebook.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "import json\n",
    "import os\n",
    "import _jsonnet\n",
    "import os"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "from seq2struct.commands.infer import Inferer\n",
    "from seq2struct.datasets.spider import SpiderItem\n",
    "from seq2struct.utils import registry"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "import torch"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "exp_config = json.loads(\n",
    "    _jsonnet.evaluate_file(\n",
    "        \"experiments/spider-configs/gap-run.jsonnet\"))\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "model_config_path = exp_config[\"model_config\"]\n",
    "model_config_args = exp_config.get(\"model_config_args\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "infer_config = json.loads(\n",
    "    _jsonnet.evaluate_file(\n",
    "        model_config_path, \n",
    "        tla_codes={'args': json.dumps(model_config_args)}))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "infer_config[\"model\"][\"encoder_preproc\"][\"db_path\"] = \"data/sqlite_files/\""
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "inferer = Inferer(infer_config)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "model_dir = exp_config[\"logdir\"] + \"/bs=12,lr=1.0e-04,bert_lr=1.0e-05,end_lr=0e0,att=1\"\n",
    "checkpoint_step = exp_config[\"eval_steps\"][0]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "model = inferer.load_model(model_dir, checkpoint_step)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "from seq2struct.datasets.spider_lib.preprocess.get_tables import dump_db_json_schema\n",
    "from seq2struct.datasets.spider import load_tables_from_schema_dict"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "db_id = \"singer\""
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "my_schema = dump_db_json_schema(\"data/sqlite_files/{db_id}/{db_id}.sqlite\".format(db_id=db_id), db_id)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "from seq2struct.utils.api_utils import refine_schema_names"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "my_schema"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "# If you want to change your schema name, then run this; Otherwise you can skip this.\n",
    "refine_schema_names(my_schema)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "schema, eval_foreign_key_maps = load_tables_from_schema_dict(my_schema)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "schema.keys()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "dataset = registry.construct('dataset_infer', {\n",
    "   \"name\": \"spider\", \"schemas\": schema, \"eval_foreign_key_maps\": eval_foreign_key_maps, \n",
    "    \"db_path\": \"data/sqlite_files/\"\n",
    "})"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "for _, schema in dataset.schemas.items():\n",
    "    model.preproc.enc_preproc._preprocess_schema(schema)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "spider_schema = dataset.schemas[db_id]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "def infer(question):\n",
    "    data_item = SpiderItem(\n",
    "            text=None,  # intentionally None -- should be ignored when the tokenizer is set correctly\n",
    "            code=None,\n",
    "            schema=spider_schema,\n",
    "            orig_schema=spider_schema.orig,\n",
    "            orig={\"question\": question}\n",
    "        )\n",
    "    model.preproc.clear_items()\n",
    "    enc_input = model.preproc.enc_preproc.preprocess_item(data_item, None)\n",
    "    preproc_data = enc_input, None\n",
    "    with torch.no_grad():\n",
    "        output = inferer._infer_one(model, data_item, preproc_data, beam_size=1, use_heuristic=True)\n",
    "    return output[0][\"inferred_code\"]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "code = infer(\"How many singers are there?\")\n",
    "print(code)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3",
   "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": "3.7.9"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 4
}