docs: fixing the active learning tutorial with small-text (#1726)

* docs: fixing the active learning tutorial with `small-text`

* docs: using a tiny model

* docs: Change tutorial title

* docs: Change active learning title in card

(cherry picked from commit f4f2289)

Closes #1693

docs/tutorials/active-learning.md

@@ -11,7 +11,7 @@ These tutorials show you how to create Active Learning workflows with Rubrix.
 Build an active learning prototype with Rubrix, ModAL and scikit-learn to filter spam from the YouTube Spam Collection dataset.
 ```
 
-```{grid-item-card} 👂 Learn actively, and listen carefully to small-text
+```{grid-item-card} 👂 Active learning for text classification with small-text
 :img-top: ../_static/tutorials/active_learning_with_small_text/screenshot.png
 :link: active_learning_with_small_text.ipynb
 

docs/tutorials/active_learning_with_small_text.ipynb

@@ -5,7 +5,7 @@
    "id": "e7b5a1fa-9265-4fb8-9c0e-a9156c22f4ff",
    "metadata": {},
    "source": [
-    "# 👂 Learn actively, and listen carefully to small-text"
+    "# 👂 Active learning for text classification with small-text"
    ]
   },
   {
@@ -111,7 +111,7 @@
    "source": [
     "import datasets\n",
     "\n",
-    "trec = datasets.load_dataset('trec')"
+    "trec = datasets.load_dataset('trec', revision=\"bc790b9ce61d4c2b1ea9622cd65da40182725a61\")"
    ]
   },
   {
@@ -143,7 +143,7 @@
     "from transformers import AutoTokenizer\n",
     "\n",
     "# Choose transformer model\n",
-    "TRANSFORMER_MODEL = \"google/electra-small-discriminator\"\n",
+    "TRANSFORMER_MODEL = \"prajjwal1/bert-tiny\"\n",
     "\n",
     "# Init tokenizer\n",
     "tokenizer = AutoTokenizer.from_pretrained(TRANSFORMER_MODEL)\n",
@@ -173,15 +173,14 @@
    "outputs": [],
    "source": [
     "from small_text.integrations.transformers import TransformersDataset\n",
-    "from small_text.base import LABEL_UNLABELED\n",
     "\n",
     "# Set convenient output format \n",
     "trec_tokenized.set_format(\"torch\")\n",
     "\n",
     "# Create tuples from the tokenized training data\n",
     "data = [\n",
     "    # Need to add an extra dimension to indicate a batch size of 1 -> [None]\n",
-    "    (row[\"input_ids\"][None], row[\"attention_mask\"][None], LABEL_UNLABELED) \n",
+    "    (row[\"input_ids\"][None], row[\"attention_mask\"][None], int(row[\"coarse_label\"])) \n",
     "    for row in trec_tokenized[\"train\"]\n",
     "]\n",
     "\n",
@@ -206,7 +205,7 @@
    "source": [
     "# Create test dataset\n",
     "data_test = [\n",
-    "    (row[\"input_ids\"][None], row[\"attention_mask\"][None], int(row[\"label-coarse\"])) \n",
+    "    (row[\"input_ids\"][None], row[\"attention_mask\"][None], int(row[\"coarse_label\"])) \n",
     "    for row in trec_tokenized[\"test\"]\n",
     "]\n",
     "dataset_test = TransformersDataset(data_test)"
@@ -254,7 +253,7 @@
     "    num_classes=6,\n",
     "    # If you have a cuda device, specify it here.\n",
     "    # Otherwise, just remove the following line.\n",
-    "    kwargs={\"device\": \"cuda\"}\n",
+    "    # kwargs={\"device\": \"cuda\"}\n",
     ")\n",
     "\n",
     "# Define our query strategy\n",
@@ -319,12 +318,11 @@
    "source": [
     "import rubrix as rb\n",
     "\n",
-    "\n",
     "# Choose a name for the dataset\n",
     "DATASET_NAME = \"trec_with_active_learning\"\n",
     "\n",
     "# Define labeling schema\n",
-    "labels = trec[\"train\"].features[\"label-coarse\"].names\n",
+    "labels = trec[\"train\"].features[\"coarse_label\"].names\n",
     "settings = rb.TextClassificationSettings(label_schema=labels)\n",
     "\n",
     "# Create dataset with a label schema\n",
@@ -367,7 +365,7 @@
     "from sklearn.metrics import accuracy_score\n",
     "\n",
     "# Define some helper variables\n",
-    "LABEL2INT = trec[\"train\"].features[\"label-coarse\"].str2int\n",
+    "LABEL2INT = trec[\"train\"].features[\"coarse_label\"].str2int\n",
     "ACCURACIES = []\n",
     "\n",
     "# Set up the active learning loop with the listener decorator\n",
@@ -396,11 +394,11 @@
     "    # 2. Query active learner\n",
     "    print(\"Querying new data points ...\")\n",
     "    queried_indices = active_learner.query(num_samples=NUM_SAMPLES)\n",
-    "    ctx.query_params[\"batch_id\"] += 1\n",
+    "    new_batch = ctx.query_params[\"batch_id\"] + 1\n",
     "    new_records = [\n",
     "        rb.TextClassificationRecord(\n",
     "            text=trec[\"train\"][\"text\"][idx], \n",
-    "            metadata={\"batch_id\": ctx.query_params[\"batch_id\"]},\n",
+    "            metadata={\"batch_id\": new_batch},\n",
     "            id=idx,\n",
     "        ) \n",
     "        for idx in queried_indices\n",
@@ -415,9 +413,11 @@
     "        dataset_test.y, \n",
     "        active_learner.classifier.predict(dataset_test),\n",
     "    )\n",
+    "    \n",
     "    ACCURACIES.append(accuracy)\n",
+    "    ctx.query_params[\"batch_id\"] = new_batch\n",
     "    print(\"Done!\")\n",
-    "    \n",
+    "\n",
     "    print(\"Waiting for annotations ...\")"
    ]
   },