Upload folder using huggingface_hub

learn_with_history_visualisation.ipynb

@@ -367,41 +367,22 @@
    "id": "fe8ce873",
    "metadata": {},
    "source": [
+    "Confusion Matrix Analysis\n",
+    "An examination of the confusion matrix demonstrates that the proposed classifier exhibits robust performance, particularly considering the inherent complexity of a 31-class document categorization task. The prominent diagonal line signifies a high degree of correlation between the ground truth and the model's predictions across the majority of categories.\n",
     "\n",
-    "Analyzing your Confusion Matrix, it's clear that the classifier is performing remarkably well given the complexity of having 31 different document categories. The strong diagonal line indicates that for most classes, the model is predicting the document type correctly.\n",
+    "Key Findings:\n",
     "\n",
-    "Key Takeaways from the Matrix:\n",
-    "Best Performing Categories: Classes like contract (12 correct), educationdocument (8 correct), taxdocument (8 correct), and invoice (7 correct) show high accuracy with very few misclassifications.\n",
+    "High-Performing Categories: The model demonstrates superior discriminative capabilities for classes such as contract (12 correct), educationdocument (8), taxdocument (8), and invoice (7). These categories show high classification accuracy with negligible misclassification rates, suggesting the model has successfully captured their distinct structural or linguistic features.\n",
     "\n",
-    "Minor Confusions:\n",
+    "Inter-class Ambiguities:\n",
     "\n",
-    "courtdocument is occasionally mistaken for contract (1 instance). This makes sense as legal language can be very similar.\n",
+    "A minor degree of confusion was observed between courtdocument and contract. This overlap is conceptually justified, as both categories frequently employ specialized legal terminology and formal syntactical structures, leading to high lexical similarity.\n",
     "\n",
-    "idcard was once confused with cv, likely due to both containing personal names and profile-like information.\n",
+    "The misclassification of an idcard as a cv suggests that the model may be responding to shared attributes, specifically the presence of personal identifiers and profile-oriented information layouts.\n",
     "\n",
-    "medicaldocument seems to be a slight \"magnet\" for errors; documents like vaccinationcard and medicaldocument itself (predicted as referral) show some overlap in medical terminology.\n",
+    "The medicaldocument class acts as a slight focal point for errors regarding related sub-categories (e.g., vaccinationcard and referral). This indicates a high degree of semantic overlap in medical terminology, which presents a challenge for fine-grained classification.\n",
     "\n",
-    "Data Sparsity: Many categories (like bankstatement or birthcertificate) have very few samples in this validation set (only 1). While the model got them right, more data would be needed to confirm its stability for these specific types."
-   ]
-  },
-  {
-   "cell_type": "code",
-   "execution_count": null,
-   "id": "0d159f73",
-   "metadata": {},
-   "outputs": [],
-   "source": [
-    "from huggingface_hub import HfApi\n",
-    "\n",
-    "api = HfApi()\n",
-    "\n",
-    "# Wysyłanie całego folderu z nową wersją\n",
-    "api.upload_folder(\n",
-    "    folder_path=\"../\",\n",
-    "    repo_id=\"twoja_nazwa/nazwa-modelu\",\n",
-    "    repo_type=\"model\", # lub \"dataset\"\n",
-    "    commit_message=\"Aktualizacja modelu v2\"\n",
-    ")"
+    "Data Sparsity and Generalization: Several categories, including bankstatement and birthcertificate, are represented by a limited number of samples within the validation set. While the model correctly identified these instances, the statistical significance of these results remains constrained. Further validation utilizing a more balanced and extensive dataset is required to confirm the model’s stability and generalizability across these sparsely represented classes."
    ]
   }
  ],