Create README.md

README.md

@@ -0,0 +1,80 @@
+---
+license: apache-2.0
+language:
+- en
+---
+This is a Re-act style model.
+
+Dataset was parsed with:
+```
+def extract_trajectory_info(data):
+    """
+    Extracts the question, thoughts, actions, and observations from the trajectory field of the data.
+
+    Parameters:
+    data (dict): The data entry containing the trajectory field.
+
+    Returns:
+    dict: A dictionary containing the extracted question, thoughts, actions, and observations.
+    """
+    # Extracting the question
+    question = data.get('question', '')
+
+    # Extracting thoughts, actions, and observations using regex
+    thoughts = re.findall(r'Thought \d+: (.+?)(?=Action|\Z)', data.get('trajectory', ''), re.DOTALL)
+    actions = re.findall(r'Action \d+: (.+?)(?=Observation|\Z)', data.get('trajectory', ''), re.DOTALL)
+    observations = re.findall(r'Observation \d+: (.+?)(?=Thought|\Z)', data.get('trajectory', ''), re.DOTALL)
+
+    # Cleaning up the extracted data
+    thoughts = [thought.strip() for thought in thoughts]
+    actions = [action.strip() for action in actions]
+    observations = [observation.strip() for observation in observations]
+
+    return {
+        'question': question,
+        'thoughts': thoughts,
+        'actions': actions,
+        'observations': observations
+    }
+# Sample data
+extracted_info = extract_trajectory_info(ds["train"][0])
+```
+Then remade into a new dataset with
+```
+# Predefine the instructions for the task
+preamble = """Tools available:
+(1) Search[entity], which searches the exact entity on Wikipedia and returns the first paragraph if it exists. If not, it will return some similar entities to search.
+(2) Lookup[keyword], which returns the next sentence containing the keyword in the current passage.
+(3) Finish[answer], which returns the answer and finishes the task.
+"""
+dataset = []
+# Iterate through a specified number of examples in the training set
+for i in range(len(ds['train'])):
+    extracted_info = extract_trajectory_info(ds['train'][i])
+
+    # Iterate through each thought in the extracted information
+    for j in range(len(extracted_info['thoughts'])):
+        out = f"{preamble}---\nQuestion: {extracted_info['question']}\n"
+        prev = ""
+        # Construct output for the first thought
+        if j == 0:
+            out += f"Thought: {extracted_info['thoughts'][0]}\n"
+            out += f"Action: {extracted_info['actions'][0]}\nPAUSE\n\n\n\n"
+
+        else:
+            for k in range(1, j + 1):
+                # Use appropriate indexing to avoid out-of-bounds errors
+                prev += f"Thought:{extracted_info['thoughts'][j - k]}\n"
+                prev += f"Action: {extracted_info['actions'][j - k]}\nPAUSE\n"
+
+                prev += f"Observation: {extracted_info['observations'][j - k]}\n"
+
+            out += prev  # Remove trailing space
+            out += f"---\nThought: {extracted_info['thoughts'][j]}\n"
+            out += f"Action: {extracted_info['actions'][j]}\nPAUSE\n\n\n\n"
+
+        # Print the constructed output
+        print(out)
+        dataset.append(out)
+        #print(len(out))
+```