Add/tool access fix

README.md

@@ -70,17 +70,9 @@ Analysis and interpretation are documented within individual experiment director
 If you use MAFBench in academic work, please cite:
 
 ```bibtex
-@article{orogat2026mafbench,
-  title={Understanding Multi-Agent LLM Frameworks: A Unified Benchmark and Experimental Analysis},
-  author={Orogat, Abdelghny and Rostam, Ana and Mansour, Essam},
-  eprint={2602.03128},
-  archivePrefix={arXiv},
-  url={https://arxiv.org/abs/2602.03128}, 
-  year={2026}
-}
+
 ```
 
 ## Contact
 
-Abdelghny Orogat — Concordia University  
-Email: Abdelghny.Orogat@concordia.ca
+

single_agent/specialization/EU-IT/dataset_profile_tool/euit_data_scientist.json

@@ -0,0 +1,26 @@
+{
+  "dataset": "EU-IT",
+  "dataset_file": "EU-IT_cleaned.csv",
+  "target_column": "Position",
+  "role": "data_scientist",
+  "condition": "tool_access",
+  "output_file": "specialization_runs/tool_access/euit_data_scientist.py",
+  "prompt": "Load the dataset EU-IT_cleaned.csv from the current working directory. The dataset contains a target column named Position.\nYour job is to apply data-scientist expertise practices:\n 1) Profile data: verify delimiter, inspect column types, detect categorical vs numerical features, and check for missing values.\n 2) Clean data: Handle missing inputs and invalid values (drop the rows).\n 3) Engineer features: impute missing numeric values with the mean, impute categorical values with the most frequent value,\n      encode categorical using OneHotEncoder or LabelEncoder, and scale numerical features.\n 4) Model: train a multiclass classifier (RandomForestClassifier, GradientBoostingClassifier, or multinomial LogisticRegression) to predict the target column.\n 5) Evaluate: Perform the 80/20 train/test split and then compute Accuracy, Precision, Recall, f1-score and print a classification report.\n              Use zero_division=0 in precision/recall/f1 to prevent UndefinedMetricWarning.\n\nYou have access to a tool named 'dataset_profile' that returns factual information about the EU-IT dataset. You may call it if useful. No additional procedural guidance is provided beyond the task definition above.",
+  "tool_available": true,
+  "dry_run": false,
+  "tool_called": true,
+  "tool_call_count": 1,
+  "tool_calls": [
+    {
+      "dataset_name": "EU-IT"
+    }
+  ],
+  "token_usage": {
+    "total_tokens": 10740,
+    "prompt_tokens": 7126,
+    "cached_prompt_tokens": 0,
+    "completion_tokens": 3614,
+    "successful_requests": 2
+  },
+  "crew_output_preview": "```python\nimport pandas as pd\nimport numpy as np\nfrom sklearn.model_selection import train_test_split\nfrom sklearn.preprocessing import StandardScaler, LabelEncoder, OneHotEncoder\nfrom sklearn.ensemble import RandomForestClassifier\nfrom sklearn.metrics import accuracy_score, precision_score, recall_score, f1_score, classification_report\nfrom sklearn.compose import ColumnTransformer\nfrom sklearn.pipeline import Pipeline\nimport warnings\nwarnings.filterwarnings('ignore')\n\n# ============================================================================\n# 1. LOAD DATA\n# ============================================================================\nprint(\"=\"*80)\nprint(\"STEP 1: LOADING DATA\")\nprint(\"=\"*80)\n\ndf = pd.read_csv('EU-IT_cleaned.csv')\nprint(f\"Dataset loaded: {df.shape[0]} rows, {df.shape[1]} columns\")\nprint(f\"\\nFirst few rows:\")\nprint(df.head())\n\n# ============================================================================\n# 2. IDENTIFY TARGET AND SEPARATE FEATURES\n# =================="
+}

single_agent/specialization/EU-IT/dataset_profile_tool/euit_data_scientist.py

@@ -0,0 +1,324 @@
+
+import pandas as pd
+import numpy as np
+from sklearn.model_selection import train_test_split
+from sklearn.preprocessing import StandardScaler, LabelEncoder, OneHotEncoder
+from sklearn.ensemble import RandomForestClassifier
+from sklearn.metrics import accuracy_score, precision_score, recall_score, f1_score, classification_report
+from sklearn.compose import ColumnTransformer
+from sklearn.pipeline import Pipeline
+import warnings
+warnings.filterwarnings('ignore')
+
+# ============================================================================
+# 1. LOAD DATA
+# ============================================================================
+print("="*80)
+print("STEP 1: LOADING DATA")
+print("="*80)
+
+df = pd.read_csv('EU-IT_cleaned.csv')
+print(f"Dataset loaded: {df.shape[0]} rows, {df.shape[1]} columns")
+print(f"\nFirst few rows:")
+print(df.head())
+
+# ============================================================================
+# 2. IDENTIFY TARGET AND SEPARATE FEATURES
+# ============================================================================
+print("\n" + "="*80)
+print("STEP 2: PROFILING DATA")
+print("="*80)
+
+target_col = 'Position'
+print(f"\nTarget column: {target_col}")
+print(f"Target value counts:\n{df[target_col].value_counts().head(10)}")
+print(f"Missing values in target: {df[target_col].isna().sum()}")
+
+print(f"\nMissing values overview:")
+missing_info = pd.DataFrame({
+    'Column': df.columns,
+    'Missing_Count': df.isnull().sum(),
+    'Missing_Percentage': (df.isnull().sum() / len(df) * 100).round(2)
+})
+print(missing_info[missing_info['Missing_Count'] > 0].sort_values('Missing_Count', ascending=False))
+
+# ============================================================================
+# 3. CLEAN DATA - DROP ROWS WITH MISSING TARGET
+# ============================================================================
+print("\n" + "="*80)
+print("STEP 3: CLEANING DATA")
+print("="*80)
+
+initial_rows = len(df)
+df = df.dropna(subset=[target_col])
+print(f"Removed rows with missing target: {initial_rows - len(df)} rows")
+print(f"Remaining rows: {len(df)}")
+
+# Separate target and features
+y = df[target_col].copy()
+X = df.drop(columns=[target_col]).copy()
+
+print(f"\nFeature matrix shape: {X.shape}")
+print(f"Target shape: {y.shape}")
+
+# ============================================================================
+# 4. IDENTIFY FEATURE TYPES
+# ============================================================================
+print("\n" + "="*80)
+print("STEP 4: IDENTIFYING FEATURE TYPES")
+print("="*80)
+
+numeric_features = X.select_dtypes(include=['int64', 'float64']).columns.tolist()
+categorical_features = X.select_dtypes(include=['object']).columns.tolist()
+
+print(f"Numeric features ({len(numeric_features)}): {numeric_features}")
+print(f"\nCategorical features ({len(categorical_features)}): {categorical_features[:10]}...")
+
+# ============================================================================
+# 5. FEATURE ENGINEERING AND PREPROCESSING
+# ============================================================================
+print("\n" + "="*80)
+print("STEP 5: FEATURE ENGINEERING & PREPROCESSING")
+print("="*80)
+
+# Create copies for preprocessing
+X_processed = X.copy()
+
+# Drop Timestamp because its near-unique values add noise rather than signal.
+if 'Timestamp' in X_processed.columns:
+    X_processed = X_processed.drop(columns=['Timestamp'])
+    categorical_features = [col for col in categorical_features if col != 'Timestamp']
+    print("\nDropped 'Timestamp' column (high cardinality, low predictive value)")
+
+# Handle numeric features - impute with mean
+print("\nHandling numeric features:")
+for col in numeric_features:
+    missing_count = X_processed[col].isna().sum()
+    if missing_count > 0:
+        mean_val = X_processed[col].mean()
+        X_processed[col].fillna(mean_val, inplace=True)
+        print(f"  {col}: imputed {missing_count} missing values with mean {mean_val:.2f}")
+    else:
+        print(f"  {col}: no missing values")
+
+# Handle categorical features - impute with most frequent value
+print("\nHandling categorical features:")
+for col in categorical_features:
+    missing_count = X_processed[col].isna().sum()
+    if missing_count > 0:
+        mode_val = X_processed[col].mode()[0] if len(X_processed[col].mode()) > 0 else 'Unknown'
+        X_processed[col].fillna(mode_val, inplace=True)
+        print(f"  {col}: imputed {missing_count} missing values with mode '{mode_val}'")
+    else:
+        print(f"  {col}: no missing values")
+
+# Remove rows with any remaining NaN values
+X_processed = X_processed.dropna()
+y = y[X_processed.index]
+
+print(f"\nAfter imputation and cleaning: {X_processed.shape}")
+
+# Group singleton categories before encoding so rare labels do not dominate.
+for col in categorical_features:
+    value_counts = X_processed[col].value_counts()
+    rare_categories = value_counts[value_counts < 2].index
+    if len(rare_categories) > 0:
+        X_processed[col] = X_processed[col].replace(rare_categories, 'Other')
+
+# Identify low-cardinality categorical features for one-hot encoding
+print("\nCategorical feature cardinality:")
+low_cardinality_cats = []
+high_cardinality_cats = []
+
+for col in categorical_features:
+    cardinality = X_processed[col].nunique()
+    if cardinality <= 20:
+        low_cardinality_cats.append(col)
+        print(f"  {col}: {cardinality} unique values (ONE-HOT)")
+    else:
+        high_cardinality_cats.append(col)
+        print(f"  {col}: {cardinality} unique values (LABEL)")
+
+# ============================================================================
+# 6. ENCODE CATEGORICAL FEATURES
+# ============================================================================
+print("\n" + "="*80)
+print("STEP 6: ENCODING CATEGORICAL FEATURES")
+print("="*80)
+
+# One-hot encode low-cardinality categorical features
+if low_cardinality_cats:
+    print(f"\nOne-hot encoding: {low_cardinality_cats}")
+    X_onehot = pd.get_dummies(X_processed[low_cardinality_cats], drop_first=True)
+    X_processed = X_processed.drop(columns=low_cardinality_cats)
+    X_processed = pd.concat([X_processed, X_onehot], axis=1)
+    print(f"After one-hot encoding: {X_processed.shape[1]} features")
+
+# Label encode high-cardinality categorical features
+if high_cardinality_cats:
+    print(f"\nLabel encoding: {high_cardinality_cats}")
+    label_encoders = {}
+    for col in high_cardinality_cats:
+        le = LabelEncoder()
+        X_processed[col] = le.fit_transform(X_processed[col].astype(str))
+        label_encoders[col] = le
+    print(f"After label encoding: {X_processed.shape[1]} features")
+
+print(f"\nFinal feature matrix shape: {X_processed.shape}")
+print(f"Features: {X_processed.columns.tolist()[:10]}...")
+
+# ============================================================================
+# 7. SCALE NUMERICAL FEATURES
+# ============================================================================
+print("\n" + "="*80)
+print("STEP 7: SCALING NUMERICAL FEATURES")
+print("="*80)
+
+scaler = StandardScaler()
+numeric_cols_to_scale = numeric_features.copy()
+X_processed[numeric_cols_to_scale] = scaler.fit_transform(X_processed[numeric_cols_to_scale])
+print(f"Scaled {len(numeric_cols_to_scale)} numeric features")
+print(f"Final feature matrix shape: {X_processed.shape}")
+
+# ============================================================================
+# 8. ENCODE TARGET VARIABLE
+# ============================================================================
+print("\n" + "="*80)
+print("STEP 8: ENCODING TARGET VARIABLE")
+print("="*80)
+
+target_encoder = LabelEncoder()
+y_encoded = target_encoder.fit_transform(y)
+print(f"Target classes: {len(target_encoder.classes_)}")
+print(f"Target encoding completed")
+
+# ============================================================================
+# 9. TRAIN-TEST SPLIT (80/20)
+# ============================================================================
+print("\n" + "="*80)
+print("STEP 9: TRAIN-TEST SPLIT (80/20)")
+print("="*80)
+
+class_counts = pd.Series(y_encoded).value_counts()
+min_class_size = int(class_counts.min())
+use_stratify = min_class_size >= 2
+
+if use_stratify:
+    print("Using stratified split because every class has at least 2 samples.")
+    stratify_labels = y_encoded
+else:
+    print("Skipping stratified split because some target classes have fewer than 2 samples.")
+    print(f"Smallest class size: {min_class_size}")
+    stratify_labels = None
+
+X_train, X_test, y_train, y_test = train_test_split(
+    X_processed,
+    y_encoded,
+    test_size=0.2,
+    random_state=42,
+    stratify=stratify_labels,
+)
+
+print(f"Training set: {X_train.shape[0]} samples")
+print(f"Test set: {X_test.shape[0]} samples")
+print(f"Training set class distribution:\n{pd.Series(y_train).value_counts().head()}")
+print(f"\nTest set class distribution:\n{pd.Series(y_test).value_counts().head()}")
+
+# ============================================================================
+# 10. TRAIN MULTICLASS CLASSIFIER
+# ============================================================================
+print("\n" + "="*80)
+print("STEP 10: TRAINING MULTICLASS CLASSIFIER")
+print("="*80)
+
+model = RandomForestClassifier(
+    n_estimators=100,
+    max_depth=15,
+    min_samples_split=5,
+    min_samples_leaf=2,
+    random_state=42,
+    n_jobs=-1,
+)
+
+print("Training RandomForestClassifier...")
+model.fit(X_train, y_train)
+print("Model training completed!")
+
+# ============================================================================
+# 11. MAKE PREDICTIONS
+# ============================================================================
+print("\n" + "="*80)
+print("STEP 11: MAKING PREDICTIONS")
+print("="*80)
+
+y_pred_train = model.predict(X_train)
+y_pred_test = model.predict(X_test)
+
+print(f"Predictions generated for training set: {len(y_pred_train)}")
+print(f"Predictions generated for test set: {len(y_pred_test)}")
+
+# ============================================================================
+# 12. EVALUATE MODEL
+# ============================================================================
+print("\n" + "="*80)
+print("STEP 12: MODEL EVALUATION")
+print("="*80)
+
+# Calculate metrics
+train_accuracy = accuracy_score(y_train, y_pred_train)
+test_accuracy = accuracy_score(y_test, y_pred_test)
+
+test_precision = precision_score(y_test, y_pred_test, zero_division=0, average='weighted')
+test_recall = recall_score(y_test, y_pred_test, zero_division=0, average='weighted')
+test_f1 = f1_score(y_test, y_pred_test, zero_division=0, average='weighted')
+
+print("\n" + "-"*80)
+print("OVERALL PERFORMANCE METRICS (Test Set)")
+print("-"*80)
+print(f"Accuracy:  {test_accuracy:.4f}")
+print(f"Precision: {test_precision:.4f}")
+print(f"Recall:    {test_recall:.4f}")
+print(f"F1-Score:  {test_f1:.4f}")
+
+print("\n" + "-"*80)
+print("TRAINING vs TEST ACCURACY")
+print("-"*80)
+print(f"Training Accuracy: {train_accuracy:.4f}")
+print(f"Test Accuracy:     {test_accuracy:.4f}")
+
+# ============================================================================
+# 13. DETAILED CLASSIFICATION REPORT
+# ============================================================================
+print("\n" + "="*80)
+print("DETAILED CLASSIFICATION REPORT (Test Set)")
+print("="*80)
+
+class_names = target_encoder.classes_
+report = classification_report(
+    y_test,
+    y_pred_test,
+    labels=np.arange(len(class_names)),
+    target_names=class_names,
+    zero_division=0,
+    digits=3
+)
+print(report)
+
+# ============================================================================
+# 14. DISPLAY FIRST 10 PREDICTIONS vs ACTUAL
+# ============================================================================
+print("\n" + "="*80)
+print("FIRST 10 PREDICTIONS vs ACTUAL LABELS (Test Set)")
+print("="*80)
+
+results_df = pd.DataFrame({
+    'Predicted': target_encoder.inverse_transform(y_pred_test[:10]),
+    'Actual': target_encoder.inverse_transform(y_test[:10]),
+    'Match': y_pred_test[:10] == y_test[:10]
+})
+
+print(results_df.to_string(index=True))
+
+print("\n" + "="*80)
+print("PIPELINE EXECUTION COMPLETED SUCCESSFULLY")
+print("="*80)