Upload industrial policy classifier model (hub_ready) with automated model card

README.md

@@ -1,3 +1,169 @@
----
-license: cc-by-4.0
----
+---
+license: apache-2.0
+base_model: bert-base-uncased
+tags:
+- text-classification
+- industrial-policy
+- economics
+- policy-analysis
+- bert
+- government-policy
+- trade-policy
+language:
+- en
+pipeline_tag: text-classification
+widget:
+- text: "Government provides subsidies to promote renewable energy development"
+  example_title: "IP goal Example"
+- text: "Company announces quarterly earnings report"
+  example_title: "No IP goal Example"
+- text: "The document mentions policy changes"
+  example_title: "Not enough information Example"
+metrics:
+- accuracy
+- f1
+- precision
+- recall
+library_name: transformers
+---
+
+# Industrial Policy Classification Model
+
+This model classifies text documents to determine whether they describe industrial policy goals. It was fine-tuned from bert-base-uncased on a dataset of policy documents and measures.
+
+## Model Description
+
+This is a BERT-based text classification model trained to identify industrial policy intentions in text. The model can classify text into 3 categories:
+
+- **IP goal** (0): Text describes an industrial policy objective or intervention
+- **No IP goal** (1): Text does not describe an industrial policy objective
+- **Not enough information** (2): Insufficient information to determine policy intent
+
+
+## Intended Use
+
+This model is designed for research purposes to analyze policy documents, government measures, and related texts to identify industrial policy intentions. It can be used by:
+
+- Economics researchers studying industrial policy
+- Policy analysts examining government interventions  
+- Data scientists working with policy text classification
+- Government agencies analyzing policy effectiveness
+
+## Model Performance
+
+- **Accuracy**: 0.941
+- **F1 Score**: 0.941
+- **Precision**: 0.941
+- **Recall**: 0.941
+- **Test Loss**: 0.2886
+
+*Metrics evaluated on held-out test set*
+
+## Training Data
+
+The model was trained on annotated policy documents including:
+- Expert-annotated policy measures from multiple countries
+- Government trade and industrial policy documents
+- WTO and multilateral organization policy entries
+- Economic policy text spanning different sectors and time periods
+
+The training dataset includes documents from various income-level countries to ensure robust performance across different economic contexts.
+
+## Training Procedure
+
+### Model Architecture
+- **Base model**: bert-base-uncased
+- **Architecture**: BertForSequenceClassification
+- **Number of labels**: 3
+- **Fine-tuning approach**: Full model fine-tuning with classification head
+
+### Training Configuration
+- **Optimization**: Hyperparameter tuning using Optuna for optimal performance
+- **Data balancing**: Oversampling applied to handle class imbalance
+- **Validation strategy**: Stratified splits with income-based validation
+- **Cross-validation**: Income group validation to test generalization
+
+## Usage
+
+```python
+from transformers import AutoTokenizer, AutoModelForSequenceClassification, pipeline
+
+# Load model and tokenizer
+model_name = "industrialpolicygroup/industrialpolicy-classifier"
+tokenizer = AutoTokenizer.from_pretrained(model_name)
+model = AutoModelForSequenceClassification.from_pretrained(model_name)
+
+# Create classification pipeline
+classifier = pipeline("text-classification", 
+                     model=model, 
+                     tokenizer=tokenizer)
+
+# Example usage
+text = "Government provides subsidies to promote renewable energy development"
+result = classifier(text)
+print(result)
+
+# Expected output format:
+# [{'label': 'LABEL_0', 'score': 0.95}]
+# 
+# Label mappings:
+
+## Limitations and Bias
+
+- The model is trained primarily on English text
+- Performance may vary on policy domains not well-represented in training data
+- The model reflects the annotation guidelines and may not capture all nuances of industrial policy
+- Bias towards certain types of policy language present in training data
+- May require domain adaptation for highly specialized policy areas
+
+## Evaluation and Validation
+
+The model underwent rigorous evaluation including:
+- Standard train/validation/test splits
+- Income-based validation across country groups
+- Cross-domain evaluation on different policy types
+- Comparison with traditional machine learning baselines
+
+## Ethical Considerations
+
+This model is intended for research and analysis purposes. Users should be aware that:
+- Policy classification can have implications for economic research and policy recommendations
+- The model's outputs should be interpreted by domain experts
+- Results should be validated against human expert judgment for critical applications
+
+## Citation
+
+If you use this model in your research, please cite:
+
+```bibtex
+@article{industrialpolicy2025,
+  title={Measuring Industrial Policy Using Natural Language Processing},
+  author={Lane, Nathaniel and [Additional Authors]},
+  journal={[Journal Name]},
+  year={2025}
+}
+```
+
+## Model Details
+
+- **Developed by**: Industrial Policy Group
+- **Model type**: Text Classification (BERT-based)
+- **Language**: English
+- **License**: Apache 2.0
+- **Fine-tuned from**: bert-base-uncased
+
+## Technical Specifications
+
+- **Input**: Text (up to 512 tokens)
+- **Output**: Classification probabilities for 3 classes
+- **Framework**: PyTorch + Transformers
+- **Model size**: ~110M parameters
+
+## Contact
+
+For questions about this model or the research, please contact the Industrial Policy Group.
+
+---
+
+*Model card auto-generated on 2025-06-19 11:24:30 from model files*
+*Source model: bert-base-uncased-3_classes-finetuned_hub_ready_20250617_151525*

config.json

@@ -0,0 +1,36 @@
+{
+  "architectures": [
+    "BertForSequenceClassification"
+  ],
+  "attention_probs_dropout_prob": 0.1,
+  "classifier_dropout": null,
+  "gradient_checkpointing": false,
+  "hidden_act": "gelu",
+  "hidden_dropout_prob": 0.1,
+  "hidden_size": 768,
+  "id2label": {
+    "0": "LABEL_0",
+    "1": "LABEL_1",
+    "2": "LABEL_2"
+  },
+  "initializer_range": 0.02,
+  "intermediate_size": 3072,
+  "label2id": {
+    "LABEL_0": 0,
+    "LABEL_1": 1,
+    "LABEL_2": 2
+  },
+  "layer_norm_eps": 1e-12,
+  "max_position_embeddings": 512,
+  "model_type": "bert",
+  "num_attention_heads": 12,
+  "num_hidden_layers": 12,
+  "pad_token_id": 0,
+  "position_embedding_type": "absolute",
+  "problem_type": "single_label_classification",
+  "torch_dtype": "float32",
+  "transformers_version": "4.52.4",
+  "type_vocab_size": 2,
+  "use_cache": true,
+  "vocab_size": 30522
+}

final_test_set_result.csv

@@ -0,0 +1,2 @@
+test_loss,test_model_preparation_time,test_accuracy,test_f1,test_precision,test_recall,test_runtime,test_samples_per_second,test_steps_per_second
+0.2885808050632477,0.0013,0.9409090909090909,0.9409010797631487,0.9409677026089238,0.9409090909090909,16.0815,27.361,3.42

model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:849c193a9fa5dc6ecb572b0dcb3ca516b5ff05bafc271a3a9e801735873a818c
+size 437961724

special_tokens_map.json

@@ -0,0 +1,7 @@
+{
+  "cls_token": "[CLS]",
+  "mask_token": "[MASK]",
+  "pad_token": "[PAD]",
+  "sep_token": "[SEP]",
+  "unk_token": "[UNK]"
+}

tokenizer_config.json

@@ -0,0 +1,56 @@
+{
+  "added_tokens_decoder": {
+    "0": {
+      "content": "[PAD]",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "100": {
+      "content": "[UNK]",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "101": {
+      "content": "[CLS]",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "102": {
+      "content": "[SEP]",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "103": {
+      "content": "[MASK]",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    }
+  },
+  "clean_up_tokenization_spaces": false,
+  "cls_token": "[CLS]",
+  "do_lower_case": true,
+  "extra_special_tokens": {},
+  "mask_token": "[MASK]",
+  "model_max_length": 512,
+  "pad_token": "[PAD]",
+  "sep_token": "[SEP]",
+  "strip_accents": null,
+  "tokenize_chinese_chars": true,
+  "tokenizer_class": "BertTokenizer",
+  "unk_token": "[UNK]"
+}

training_args.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:db96411a66f890b4c49f005ff274c91c418a46159a861b66b101e7efbc3d310f
+size 4088