updated for config by dataset (#257)

* updated for config by dataset

* updated read me

eval/README.md

@@ -126,11 +126,24 @@ Options:
 - `--size`: Default dataset size (default: 100)
 - `--seed`: Default dataset seed (default: 42)
 - `--include-params`: Include all configuration parameters (default: False)
+- `--category`: Only include datasets from this category (default: None)
+
+#### Generating Config for a Specific Category
+
+To generate a configuration file containing only datasets from a specific category:
+
+```bash
+python generate_config.py --category algorithmic --output algorithmic_datasets.yaml --model "anthropic/claude-3.5-sonnet"
+```
+
+This will create a configuration file that includes only datasets in the "algorithmic" category. This is useful when you want to focus your evaluation on a specific type of reasoning tasks.
+
+Example categories include: math, arithmetic, reasoning, algorithmic, etc. The category is automatically extracted from the dataset's module name (e.g., from `reasoning_gym.math.dataset_name`, it extracts "math").
+
+You can see all available categories by running the script without the `--category` option, as it will print all categories at the end of execution.
 
 ### Running Evaluations
 
-To run evaluations:
-
 ```bash
 python eval.py --config configs/your_config.yaml
 ```

eval/generate_config.py

@@ -15,6 +15,7 @@ Options:
     --size SIZE           Default dataset size (default: 100)
     --seed SEED           Default dataset seed (default: 42)
     --include-params      Include all configuration parameters (default: False)
+    --category CATEGORY   Only include datasets from this category (default: None)
 """
 
 import argparse
@@ -35,14 +36,27 @@ def extract_category(module_name):
     return "other"
 
 
-def generate_config(model, provider, size, seed, include_params):
-    """Generate configuration with all registered datasets."""
+def generate_config(model, provider, size, seed, include_params, category=None):
+    """Generate configuration with all registered datasets.
+
+    Args:
+        model: Model name
+        provider: Provider name
+        size: Default dataset size
+        seed: Default dataset seed
+        include_params: Whether to include all configuration parameters
+        category: If specified, only include datasets from this category
+    """
     # Group datasets by category
     categories = defaultdict(list)
 
     for dataset_name, (dataset_cls, config_cls) in DATASETS.items():
         # Extract category from module name
-        category = extract_category(dataset_cls.__module__)
+        dataset_category = extract_category(dataset_cls.__module__)
+
+        # Skip if a specific category was requested and this doesn't match
+        if category and dataset_category != category:
+            continue
 
         # Create dataset entry
         dataset_entry = {"dataset": dataset_name}
@@ -62,7 +76,7 @@ def generate_config(model, provider, size, seed, include_params):
                 dataset_entry["params"] = params
 
         # Add to appropriate category
-        categories[category].append(dataset_entry)
+        categories[dataset_category].append(dataset_entry)
 
     # Create configuration structure
     config = {
@@ -90,12 +104,18 @@ def main():
     parser.add_argument("--size", type=int, default=100, help="Default dataset size")
     parser.add_argument("--seed", type=int, default=42, help="Default dataset seed")
     parser.add_argument("--include-params", action="store_true", help="Include all configuration parameters")
+    parser.add_argument("--category", help="Only include datasets from this category")
 
     args = parser.parse_args()
 
     # Generate configuration
     config = generate_config(
-        model=args.model, provider=args.provider, size=args.size, seed=args.seed, include_params=args.include_params
+        model=args.model,
+        provider=args.provider,
+        size=args.size,
+        seed=args.seed,
+        include_params=args.include_params,
+        category=args.category,
     )
 
     # Write to file