e

Commands.txt

@@ -0,0 +1,80 @@
+Conda Set Up
+$env:Path = "C:\Users\siyov\miniconda3\Scripts;C:\Users\siyov\miniconda3;C:\Users\siyov\miniconda3\Library\bin;" + $env:Path
+conda --version
+conda init
+
+conda --version
+conda activate brighter-llm
+set OPENAI_API_KEY=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
+cd C:\Users\caoca\Documents\GitHub\BRIGHTER
+cd C:\Users\siyov\Documents\GitHub\BRIGHTER
+
+
+Mock Run
+python llms.py --model_name "openai/gpt-4" --task binary --reasoning_mode self_consistency --top_k 4 --n_shot 0 --language eng --sample_size 200 --balanced --balancing_strategy strict --prompt_variant v1 --skip_ablations --error_test Passed Mock and Error
+python llms.py --model_name "openai/gpt-4" --task intensity --reasoning_mode self_consistency --top_k 4 --n_shot 0 --language eng --sample_size 200 --balanced --balancing_strategy strict --prompt_variant v1 --skip_ablations --error_test Passed Mock and Error
+
+python llms.py --model_name "openai/gpt-4" --task binary --reasoning_mode tree_of_thoughts --n_shot 0 --language eng --sample_size 200 --balanced --balancing_strategy strict --prompt_variant tree_of_thoughts --skip_ablations --error_test Passed Mock and Error
+python llms.py --model_name "openai/gpt-4" --task intensity --reasoning_mode tree_of_thoughts --n_shot 0 --language eng --sample_size 200 --balanced --balancing_strategy strict --prompt_variant tree_of_thoughts --skip_ablations --error_test Passed Mock and Error
+
+python llms.py --model_name "openai/gpt-4" --task binary --reasoning_mode self_refine --n_shot 0 --language eng --sample_size 200 --balanced --balancing_strategy strict --prompt_variant v1 --skip_ablations --error_test Passed Mock and Error
+python llms.py --model_name "openai/gpt-4" --task intensity --reasoning_mode self_refine --n_shot 0 --language eng --sample_size 200 --balanced --balancing_strategy strict --prompt_variant v1 --skip_ablations --error_test Passed Mock and Error
+
+python llms.py --model_name "openai/gpt-4" --task binary --reasoning_mode default --n_shot 0 --language eng --sample_size 200 --balanced --balancing_strategy strict --prompt_variant v1 --skip_ablations --error_test Passed Mock and Error
+python llms.py --model_name "openai/gpt-4" --task intensity --reasoning_mode default --n_shot 0 --language eng --sample_size 200 --balanced --balancing_strategy strict --prompt_variant v1 --skip_ablations --error_test Passed Mock and Error
+
+python llms.py --model_name "openai/gpt-4" --task binary --reasoning_mode rasc --top_k 4 --n_shot 0 --language eng --sample_size 200 --balanced --balancing_strategy strict --prompt_variant v1 --skip_ablations --error_test Passed Mock and Error
+python llms.py --model_name "openai/gpt-4" --task intensity --reasoning_mode rasc --top_k 4 --n_shot 0 --language eng --sample_size 200 --balanced --balancing_strategy strict  --prompt_variant v1 --skip_ablations --error_test Passed Mock and Error
+
+python llms.py --model_name "openai/gpt-4" --task binary --reasoning_mode complexity_based --n_shot 0 --language eng --sample_size 200 --balanced --balancing_strategy strict --skip_ablations --error_test Passed Mock and Error
+python llms.py --model_name "openai/gpt-4" --task intensity --reasoning_mode complexity_based --n_shot 0 --language eng --sample_size 200 --balanced --balancing_strategy strict --skip_ablations --error_test Passed Mock and Error
+
+python llms.py --model_name "openai/gpt-4" --task binary --reasoning_mode plan_and_solve --n_shot 0 --language eng --sample_size 200 --balanced --balancing_strategy strict --prompt_variant v1 --skip_ablations --error_test Passed Mock and Error
+python llms.py --model_name "openai/gpt-4" --task intensity --reasoning_mode plan_and_solve --n_shot 0 --language eng --sample_size 200 --balanced --balancing_strategy strict --prompt_variant v1 --skip_ablations --error_test Passed Mock and Error
+
+python llms.py --model_name "openai/gpt-4" --task binary --reasoning_mode rankcot --top_k 4 --n_shot 0 --language eng --sample_size 200 --balanced --balancing_strategy strict --skip_ablations --error_test Passed Mock and Error
+python llms.py --model_name "openai/gpt-4" --task intensity --reasoning_mode rankcot --top_k 4 --n_shot 0 --language eng --sample_size 200 --balanced --balancing_strategy strict --skip_ablations --error_test Passed Mock and Error
+v
+
+
+Test Run For ChatGPT
+python llms.py --model_name "openai/gpt-4.1" --task binary --reasoning_mode self_consistency --top_k 4 --n_shot 0 --language eng --sample_size 10 --balanced --balancing_strategy strict --prompt_variant v1 --skip_ablations
+python llms.py --model_name "openai/gpt-4.1" --task intensity --reasoning_mode self_consistency --top_k 4 --n_shot 0 --language eng --sample_size 20 --balanced --balancing_strategy strict --prompt_variant v1 --skip_ablations
+python llms.py --model_name "openai/gpt-4.1" --task binary --reasoning_mode self_consistency --top_k 4 --n_shot 1 --language eng --sample_size 20 --balanced --balancing_strategy strict --prompt_variant v1 --skip_ablations
+python llms.py --model_name "openai/gpt-4.1" --task intensity --reasoning_mode self_consistency --top_k 4 --n_shot 1 --language eng --sample_size 20 --balanced --balancing_strategy strict --prompt_variant v1 --skip_ablations
+
+python llms.py --model_name "openai/gpt-4.1" --task binary --reasoning_mode tree_of_thoughts --top_k 4 --n_shot 0 --language eng --sample_size 10 --balanced --balancing_strategy strict --prompt_variant tree_of_thoughts --skip_ablations --tot_steps 3 --tot_search bfs --tot_vth 0.6 --tot_value_trials 1
+python llms.py --model_name "openai/gpt-4.1" --task intensity --reasoning_mode tree_of_thoughts --top_k 4 --n_shot 0 --language eng --sample_size 20 --balanced --balancing_strategy strict --prompt_variant tree_of_thoughts --skip_ablations --tot_steps 3 --tot_search bfs --tot_vth 0.6 --tot_value_trials 1
+python llms.py --model_name "openai/gpt-4.1" --task binary --reasoning_mode tree_of_thoughts --top_k 4 --n_shot 1 --language eng --sample_size 10 --balanced --balancing_strategy strict --prompt_variant tree_of_thoughts --skip_ablations --tot_steps 3 --tot_search bfs --tot_vth 0.6 --tot_value_trials 1
+python llms.py --model_name "openai/gpt-4.1" --task intensity --reasoning_mode tree_of_thoughts --top_k 4 --n_shot 1 --language eng --sample_size 20 --balanced --balancing_strategy strict --prompt_variant tree_of_thoughts --skip_ablations --tot_steps 3 --tot_search bfs --tot_vth 0.6 --tot_value_trials 1
+
+python llms.py --model_name "openai/gpt-4.1" --task binary --reasoning_mode self_refine --n_shot 0 --language eng --sample_size 10 --balanced --balancing_strategy strict --prompt_variant v1 --skip_ablations
+python llms.py --model_name "openai/gpt-4.1" --task intensity --reasoning_mode self_refine --n_shot 0 --language eng --sample_size 20 --balanced --balancing_strategy strict --prompt_variant v1 --skip_ablations
+python llms.py --model_name "openai/gpt-4.1" --task binary --reasoning_mode self_refine --n_shot 1 --language eng --sample_size 10 --balanced --balancing_strategy strict --prompt_variant v1 --skip_ablations
+python llms.py --model_name "openai/gpt-4.1" --task intensity --reasoning_mode self_refine --n_shot 1 --language eng --sample_size 20 --balanced --balancing_strategy strict --prompt_variant v1 --skip_ablations
+
+python llms.py --model_name "openai/gpt-4.1" --task binary --reasoning_mode default --n_shot 0 --language eng --sample_size 10 --balanced --balancing_strategy strict --prompt_variant v1 --skip_ablations
+python llms.py --model_name "openai/gpt-4.1" --task intensity --reasoning_mode default --n_shot 0 --language eng --sample_size 20 --balanced --balancing_strategy strict --prompt_variant v1 --skip_ablations
+python llms.py --model_name "openai/gpt-4.1" --task binary --reasoning_mode default --n_shot 1 --language eng --sample_size 10 --balanced --balancing_strategy strict --prompt_variant v1 --skip_ablations
+python llms.py --model_name "openai/gpt-4.1" --task intensity --reasoning_mode default --n_shot 1 --language eng --sample_size 20 --balanced --balancing_strategy strict --prompt_variant v1 --skip_ablations
+
+python llms.py --model_name "openai/gpt-4.1" --task binary --reasoning_mode rasc --top_k 4 --n_shot 0 --language eng --sample_size 10 --balanced --balancing_strategy strict --prompt_variant v1 --skip_ablations
+python llms.py --model_name "openai/gpt-4.1" --task intensity --reasoning_mode rasc --top_k 4 --n_shot 0 --language eng --sample_size 20 --balanced --balancing_strategy strict --prompt_variant v1 --skip_ablations
+python llms.py --model_name "openai/gpt-4.1" --task binary --reasoning_mode rasc --top_k 4 --n_shot 1 --language eng --sample_size 10 --balanced --balancing_strategy strict --prompt_variant v1 --skip_ablations
+python llms.py --model_name "openai/gpt-4.1" --task intensity --reasoning_mode rasc --top_k 4 --n_shot 1 --language eng --sample_size 20 --balanced --balancing_strategy strict --prompt_variant v1 --skip_ablations
+
+python llms.py --model_name "openai/gpt-4.1" --task binary --reasoning_mode complexity_based --n_shot 0 --language eng --sample_size 10 --balanced --balancing_strategy strict --skip_ablations
+python llms.py --model_name "openai/gpt-4.1" --task intensity --reasoning_mode complexity_based --n_shot 0 --language eng --sample_size 20 --balanced --balancing_strategy strict --skip_ablations
+python llms.py --model_name "openai/gpt-4.1" --task binary --reasoning_mode complexity_based --n_shot 1 --language eng --sample_size 10 --balanced --balancing_strategy strict --skip_ablations
+python llms.py --model_name "openai/gpt-4.1" --task intensity --reasoning_mode complexity_based --n_shot 1 --language eng --sample_size 20 --balanced --balancing_strategy strict --skip_ablations
+
+python llms.py --model_name "openai/gpt-4.1" --task binary --reasoning_mode plan_and_solve --n_shot 0 --language eng --sample_size 10 --balanced --balancing_strategy strict --prompt_variant v1 --skip_ablations
+python llms.py --model_name "openai/gpt-4.1" --task intensity --reasoning_mode plan_and_solve --n_shot 0 --language eng --sample_size 20 --balanced --balancing_strategy strict --prompt_variant v1 --skip_ablations
+python llms.py --model_name "openai/gpt-4.1" --task binary --reasoning_mode plan_and_solve --n_shot 1 --language eng --sample_size 10 --balanced --balancing_strategy strict --prompt_variant v1 --skip_ablations
+python llms.py --model_name "openai/gpt-4.1" --task intensity --reasoning_mode plan_and_solve --n_shot 1 --language eng --sample_size 20 --balanced --balancing_strategy strict --prompt_variant v1 --skip_ablations
+
+python llms.py --model_name "openai/gpt-4.1" --task binary --reasoning_mode rankcot --top_k 4 --n_shot 0 --language eng --sample_size 10 --balanced --balancing_strategy strict --skip_ablations
+python llms.py --model_name "openai/gpt-4.1" --task intensity --reasoning_mode rankcot --top_k 4 --n_shot 0 --language eng --sample_size 20 --balanced --balancing_strategy strict --skip_ablations
+python llms.py --model_name "openai/gpt-4.1" --task binary --reasoning_mode rankcot --top_k 4 --n_shot 1 --language eng --sample_size 10 --balanced --balancing_strategy strict --skip_ablations
+python llms.py --model_name "openai/gpt-4.1" --task intensity --reasoning_mode rankcot --top_k 4 --n_shot 1 --language eng --sample_size 20 --balanced --balancing_strategy strict --skip_ablations
+v

EmotionsAI.py

@@ -0,0 +1,197 @@
+'''from pyplutchik import plutchik
+import matplotlib.pyplot as plt
+
+emotions = {
+    "joy": 1,
+
+    "fear": 0.2,
+    "surprise": 0.3,
+    "sadness": 0.7,
+    "disgust": 0.1,
+    "anger": 0.8,
+
+}
+
+plutchik(emotions)
+plt.show()  # <-- this line makes the window pop up!
+'''
+'''import pandas as pd
+
+# Choose your language code (e.g., 'arq', 'hin', 'eng' — only 10 have intensities)
+lang = 'eng'
+
+# File paths
+categories_base = "hf://datasets/brighter-dataset/BRIGHTER-emotion-categories/"
+intensities_base = "hf://datasets/brighter-dataset/BRIGHTER-emotion-intensities/"
+
+splits = {
+    'train': f"{lang}/train-00000-of-00001.parquet",
+    'dev': f"{lang}/dev-00000-of-00001.parquet",
+    'test': f"{lang}/test-00000-of-00001.parquet"
+}
+
+# Load both category and intensity splits
+cat_df = pd.read_parquet(categories_base + splits["train"])
+int_df = pd.read_parquet(intensities_base + splits["train"])
+
+# Peek at both to check structure
+print("Categories sample:\n", cat_df.head())
+print("Intensities sample:\n", int_df.head())'''
+
+# Required installations:
+# pip install openai pandas scikit-learn scipy pyarrow fsspec huggingface_hub
+
+import os
+import openai
+import pandas as pd
+import json
+from collections import Counter, defaultdict
+from sklearn.metrics import f1_score
+from scipy.stats import pearsonr
+
+# Set your OpenAI API key
+openai.api_key = "sk-..."  # Replace with your GPT-4 key
+
+EMOTIONS = ["anger", "disgust", "fear", "joy", "sadness", "surprise"]
+
+# Few-shot examples with CoT-style reasoning
+FEW_SHOT_EXAMPLES = """
+Example 1:
+Text: "I finally got the job I’ve been dreaming about!"
+Reasoning: The speaker expresses excitement and satisfaction. This suggests joy and surprise.
+Emotions: ["joy", "surprise"]
+Intensity: {"joy": 3, "surprise": 2}
+
+Example 2:
+Text: "She betrayed me after everything we went through."
+Reasoning: Betrayal often evokes anger and sadness.
+Emotions: ["anger", "sadness"]
+Intensity: {"anger": 2, "sadness": 3}
+
+Example 3:
+Text: "The room was silent. I could feel my heart pounding faster with each second."
+Reasoning: This situation evokes tension and uncertainty, associated with fear.
+Emotions: ["fear"]
+Intensity: {"fear": 2}
+"""
+
+# Prompt with few-shot + step-by-step
+PROMPT_TEMPLATE = FEW_SHOT_EXAMPLES + """
+
+Now analyze the following:
+Text: "{text}"
+Think step by step and explain your reasoning. Then output your final answer in this format:
+Emotions: ["emotion1", "emotion2", ...]
+Intensity: {"emotion1": x, "emotion2": x, ...}
+"""
+'''
+# Parse model response
+def parse_response(response_text):
+    try:
+        lines = response_text.strip().splitlines()
+        emotions = json.loads(lines[-2].split(":", 1)[1].strip())
+        intensity = json.loads(lines[-1].split(":", 1)[1].strip())
+        return emotions, intensity
+    except Exception as e:
+        return None, None
+
+# Run multiple prompts and aggregate (self-consistency)
+def self_consistent_prediction(text, num_samples=5, model="gpt-4"):
+    results = []
+    for _ in range(num_samples):
+        prompt = PROMPT_TEMPLATE.format(text=text)
+        try:
+            response = openai.ChatCompletion.create(
+                model=model,
+                messages=[{"role": "user", "content": prompt}],
+                temperature=0.7
+            )
+            content = response['choices'][0]['message']['content']
+            emotions, intensity = parse_response(content)
+            if emotions and intensity:
+                results.append((emotions, intensity))
+        except Exception as e:
+            print(f"OpenAI error: {e}")
+    return aggregate_predictions(results)
+
+# Aggregate responses using majority vote (labels) and mean (intensity)
+def aggregate_predictions(results):
+    emotion_votes = Counter()
+    intensity_scores = defaultdict(list)
+
+    for emotions, intensity in results:
+        for e in emotions:
+            emotion_votes[e] += 1
+            intensity_scores[e].append(intensity.get(e, 0))
+
+    total = len(results)
+    final_emotions = [e for e, count in emotion_votes.items() if count >= total / 2]
+    final_intensity = {e: round(sum(intensity_scores[e]) / len(intensity_scores[e]), 2)
+                       for e in final_emotions}
+    return final_emotions, final_intensity
+
+# Load BRIGHTER data from Hugging Face
+def load_brighter_data(lang='arq', split='test'):
+    cat_path = f"hf://datasets/brighter-dataset/BRIGHTER-emotion-categories/{lang}/{split}-00000-of-00001.parquet"
+    int_path = f"hf://datasets/brighter-dataset/BRIGHTER-emotion-intensities/{lang}/{split}-00000-of-00001.parquet"
+    cat_df = pd.read_parquet(cat_path)
+    int_df = pd.read_parquet(int_path)
+    merged = pd.merge(cat_df, int_df, on="text", suffixes=("_cat", "_int"))
+    return merged
+
+# Convert label list to binary vector
+def labels_to_vec(label_list):
+    return [1 if e in label_list else 0 for e in EMOTIONS]
+
+# Evaluation
+def evaluate(predictions, golds, pred_intensities, gold_intensities):
+    pred_labels = [labels_to_vec(p) for p in predictions]
+    gold_labels = [labels_to_vec(g) for g in golds]
+
+    # Multi-label F1
+    f1 = f1_score(gold_labels, pred_labels, average="macro")
+
+    # Pearson per emotion
+    pearson_scores = {}
+    for e in EMOTIONS:
+        gold_vals = [gi.get(e, 0) for gi in gold_intensities]
+        pred_vals = [pi.get(e, 0) for pi in pred_intensities]
+        if any(gold_vals):
+            r, _ = pearsonr(gold_vals, pred_vals)
+            pearson_scores[e] = round(r, 3)
+        else:
+            pearson_scores[e] = "N/A"
+
+    return f1, pearson_scores
+
+# Main pipeline
+if __name__ == "__main__":
+    lang = "arq"  # any of the 10 BRIGHTER languages with intensity labels
+    df = load_brighter_data(lang=lang, split="test")
+
+    sample_size = 10  # Small batch for quick testing — increase this
+    predictions = []
+    golds = []
+    pred_intensities = []
+    gold_intensities = []
+
+    for _, row in df.sample(sample_size, random_state=42).iterrows():
+        text = row["text"]
+        gold_labels = row["labels_cat"]
+        gold_scores = row["labels_int"]
+
+        pred_labels, pred_scores = self_consistent_prediction(text, num_samples=5)
+
+        predictions.append(pred_labels)
+        pred_intensities.append(pred_scores)
+        golds.append(gold_labels)
+        gold_intensities.append(gold_scores)
+
+        print(f"\n📝 Text: {text}")
+        print(f"✅ Gold: {gold_labels} — {gold_scores}")
+        print(f"🤖 Pred: {pred_labels} — {pred_scores}")
+
+    # Final evaluation
+    f1_macro, pearson = evaluate(predictions, golds, pred_intensities, gold_intensities)
+    print("\n📊 F1 (macro):", round(f1_macro, 4))
+    print("📈 Pearson per emotion:", pearson)'''