Update structure of output folder (#27)

* Rename output directories

* Add custom warnings

* Fix key bug

* Consolidate column outputs

* Update docs

* Output columns are now fixed

* Split db into qualitative and quantitative outputs

* Re-iterate for the LLM to focus on feature

* Minor time tracking update

* Add new utility function

* Add `JurisdictionUpdater`

* Record jurisdiction info

* File name now contains download date

* Add `num_ordinances_in_doc` func

* Use `num_ordinances_in_doc` func

* Add explicit ignore to prompt

* Pass tech down one function

* Write out run meta

* Add total run time to jurisdiction info

* Process now writes time to jurisdiction file and only usage goes to usage file

* Add subdivision info

* Datetime recorded in UTC

* Updates to meta dict

* Rename dirs args

* Minor bug fix

* Track LLM parse args in meta

Author: ppinchuk

compass/extraction/solar/parse.py

@@ -4,6 +4,7 @@
 import logging
 from copy import deepcopy
 from itertools import chain
+from warnings import warn
 
 import pandas as pd
 
@@ -24,6 +25,7 @@
     setup_graph_sef_types,
     setup_multiplier,
 )
+from compass.warnings import COMPASSWarning
 
 
 logger = logging.getLogger(__name__)
@@ -169,7 +171,7 @@ async def _parse_extra_restriction(
             chat_llm_caller=self._init_chat_llm_caller(system_message),
         )
         info = await run_async_tree(tree)
-        info.update({"feature": feature})
+        info.update({"feature": feature, "quantitative": is_numerical})
         return [info]
 
     async def _parse_setback_feature(
@@ -268,7 +270,10 @@ async def _parse_p_or_np_text(
 
     async def _extract_setback_values(self, text, **kwargs):
         """Extract setback values for a given feature from input text"""
-        return await self._run_setback_graph(setup_multiplier, text, **kwargs)
+        decision_tree_out = await self._run_setback_graph(
+            setup_multiplier, text, **kwargs
+        )
+        return _update_output_keys(decision_tree_out)
 
     async def _run_setback_graph(
         self,
@@ -293,3 +298,25 @@ async def _run_setback_graph(
         if base_messages:
             return await run_async_tree_with_bm(tree, base_messages)
         return await run_async_tree(tree)
+
+
+def _update_output_keys(output):
+    """Standardize output keys
+
+    We could standardize output keys by modifying the LLM prompts, but
+    have found that it's more accurate to instruct the LLM to use
+    descriptive keys (e.g. "mult_value" instead of "value" or
+    "mult_type" instead of "units")
+    """
+
+    if "mult_value" not in output:
+        return output
+
+    output["value"] = output.pop("mult_value")
+
+    if units := output.get("units"):
+        msg = f"Found non-null units value for multiplier: {units}"
+        warn(msg, COMPASSWarning)
+    output["units"] = "structure-height-multiplier"
+
+    return output

compass/extraction/wind/graphs.py

@@ -226,6 +226,8 @@ def setup_conditional(**kwargs):
     G.add_node(
         "init",
         prompt=(
+            "Focus only on setback from {feature}; do not respond based "
+            "on any text related to {ignore_features}."
             "Does the setback from {feature} mention a minimum or maximum "
             "static setback distance regardless of the outcome of the "
             "multiplier calculation? This is often phrased as 'the greater "

compass/extraction/wind/parse.py

@@ -4,6 +4,7 @@
 import logging
 from copy import deepcopy
 from itertools import chain
+from warnings import warn
 
 import pandas as pd
 
@@ -25,6 +26,7 @@
     setup_multiplier,
     setup_conditional,
 )
+from compass.warnings import COMPASSWarning
 
 
 logger = logging.getLogger(__name__)
@@ -168,7 +170,7 @@ async def _parse_extra_restriction(
             chat_llm_caller=self._init_chat_llm_caller(system_message),
         )
         info = await run_async_tree(tree)
-        info.update({"feature": feature})
+        info.update({"feature": feature, "quantitative": is_numerical})
         return [info]
 
     async def _parse_setback_feature(
@@ -203,7 +205,7 @@ async def _base_messages(self, text, **feature_kwargs):
         """Get base messages for setback feature parsing"""
         system_message = SETBACKS_SYSTEM_MESSAGE.format(
             feature=feature_kwargs["feature"],
-            tech=feature_kwargs["wes_type"],
+            tech=feature_kwargs["tech"],
         )
         tree = setup_async_decision_tree(
             setup_base_graph,
@@ -270,8 +272,9 @@ async def _extract_setback_values(self, text, **kwargs):
         decision_tree_out = await self._run_setback_graph(
             setup_multiplier, text, **kwargs
         )
+        decision_tree_out = _update_output_keys(decision_tree_out)
 
-        if decision_tree_out.get("mult_value") is None:
+        if decision_tree_out.get("value") is None:
             return decision_tree_out
 
         decision_tree_conditional_out = await self._run_setback_graph(
@@ -303,3 +306,25 @@ async def _run_setback_graph(
         if base_messages:
             return await run_async_tree_with_bm(tree, base_messages)
         return await run_async_tree(tree)
+
+
+def _update_output_keys(output):
+    """Standardize output keys
+
+    We could standardize output keys by modifying the LLM prompts, but
+    have found that it's more accurate to instruct the LLM to use
+    descriptive keys (e.g. "mult_value" instead of "value" or
+    "mult_type" instead of "units")
+    """
+
+    if "mult_value" not in output:
+        return output
+
+    output["value"] = output.pop("mult_value")
+
+    if units := output.get("units"):
+        msg = f"Found non-null units value for multiplier: {units}"
+        warn(msg, COMPASSWarning)
+    output["units"] = output.pop("mult_type", None)
+
+    return output