utils: add notebook for generating dataset splits

CreateDatasetSplits.ipynb

@@ -0,0 +1,223 @@
+{
+ "cells": [
+  {
+   "cell_type": "code",
+   "execution_count": 1,
+   "id": "027adfa9-5e64-474b-9a95-12e5c28c90a7",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "import csv\n",
+    "import random\n",
+    "import requests"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 2,
+   "id": "e9fff288-d062-4c27-b9dc-db8579bbd3cf",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "random.seed(83607)"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 3,
+   "id": "f85d8d2c-eca9-481d-b848-4d43a072b5fb",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "# We use this as v1 of our dataset\n",
+    "revision = \"0ecb2228e6c290dd22836024f32e559cc9b9711e\"\n",
+    "original_dataset_file = \"gold_standard_v1.csv\""
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 4,
+   "id": "9c94cf8c-2185-4ca5-9191-02dd06c2fa0d",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "# Download it - the simple way\n",
+    "url = f\"https://raw.githubusercontent.com/lucijakrusic/SentiAnno/{revision}/gold_standard.csv\"\n",
+    "r = requests.get(url, allow_redirects=True)\n",
+    "\n",
+    "if r:\n",
+    "    with open(original_dataset_file, \"wb\") as f_out:\n",
+    "        f_out.write(r.content)"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 5,
+   "id": "1457fbb4-aeab-4c2e-a283-bcc12406ef3e",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "# E.g.\n",
+    "# {\"negative\": [...], \"positive\": [...]\n",
+    "label_sentences_mapping = {}\n",
+    "\n",
+    "num_examples = 0\n",
+    "\n",
+    "with open(original_dataset_file, \"rt\") as csv_file:\n",
+    "    csv_reader = csv.reader(csv_file, delimiter=',')\n",
+    "\n",
+    "    # Skip header\n",
+    "    next(csv_reader, None)\n",
+    "\n",
+    "    for line in csv_reader:\n",
+    "        assert len(line) == 5\n",
+    "\n",
+    "        sentence = line[2]\n",
+    "        label = line[-1]\n",
+    "\n",
+    "        current_example = [label, sentence]\n",
+    "        \n",
+    "        if label in label_sentences_mapping:\n",
+    "            label_sentences_mapping[label].append(current_example)\n",
+    "        else:\n",
+    "            label_sentences_mapping[label] = [current_example]\n",
+    "\n",
+    "        num_examples += 1"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 6,
+   "id": "be5eda4e-bc2b-4f24-a584-7d7b34987f73",
+   "metadata": {},
+   "outputs": [
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "Label negative has 447 sentences\n",
+      "Label mixed has 56 sentences\n",
+      "Label positive has 81 sentences\n",
+      "Label neutral has 345 sentences\n"
+     ]
+    }
+   ],
+   "source": [
+    "for label, sentences in label_sentences_mapping.items():\n",
+    "    print(\"Label\", label, \"has\", len(sentences), \"sentences\")"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 7,
+   "id": "e218106a-8f23-41b0-96e5-b57a7a83fc5f",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "# We create 80 / 10 / 10 splits, but for each label (to avoid over/under-representing labels\n",
+    "\n",
+    "train_examples = []\n",
+    "dev_examples = []\n",
+    "test_examples = []\n",
+    "\n",
+    "for _, sentences in label_sentences_mapping.items():\n",
+    "    random.shuffle(sentences)\n",
+    "\n",
+    "    split_1 = int(0.8 * len(sentences))\n",
+    "    split_2 = int(0.9 * len(sentences))\n",
+    "\n",
+    "    current_train_examples = sentences[:split_1]\n",
+    "    current_dev_examples = sentences[split_1:split_2]\n",
+    "    current_test_examples = sentences[split_2:]\n",
+    "\n",
+    "    train_examples += current_train_examples\n",
+    "    dev_examples += current_dev_examples\n",
+    "    test_examples += current_test_examples"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 8,
+   "id": "5e1fc73f-1b9b-41eb-946b-872a1308712d",
+   "metadata": {},
+   "outputs": [
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "Number of training examples: 741\n",
+      "Number of development examples: 93\n",
+      "Number of test examples: 95\n"
+     ]
+    }
+   ],
+   "source": [
+    "print(\"Number of training examples:\", len(train_examples))\n",
+    "print(\"Number of development examples:\", len(dev_examples))\n",
+    "print(\"Number of test examples:\", len(test_examples))"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 9,
+   "id": "5f359476-ecd5-4502-86ba-fee8bd8d3dcf",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "assert num_examples == (len(train_examples) + len(dev_examples) + len(test_examples))"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 14,
+   "id": "29e1f3c9-3c3b-4541-840d-3db304966762",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "def write_examples(examples: str, split_name: str):\n",
+    "    # Shuffle again for more fun ;)\n",
+    "    random.shuffle(examples)\n",
+    "    with open(f\"{split_name}.txt\", \"wt\") as f_out:\n",
+    "        for example in examples:\n",
+    "            label, sentence = example\n",
+    "\n",
+    "            # We stick to Flair format for classification tasks, which is basically FastText inspired ;)\n",
+    "            new_label = \"__label__\" + label\n",
+    "            f_out.write(f\"{new_label} {sentence}\\n\")"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 15,
+   "id": "502b3865-3efe-4730-9be8-ea675fd3feec",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "write_examples(train_examples, \"train\")\n",
+    "write_examples(dev_examples, \"dev\")\n",
+    "write_examples(test_examples, \"test\")"
+   ]
+  }
+ ],
+ "metadata": {
+  "kernelspec": {
+   "display_name": "Python 3 (ipykernel)",
+   "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": "3.12.3"
+  }
+ },
+ "nbformat": 4,
+ "nbformat_minor": 5
+}