geoai-py 0.13.0__py2.py3-none-any.whl → 0.13.2__py2.py3-none-any.whl

geoai/__init__.py

@@ -2,7 +2,7 @@
 
 __author__ = """Qiusheng Wu"""
 __email__ = "giswqs@gmail.com"
-__version__ = "0.13.0"
+__version__ = "0.13.2"
 
 
 import os

geoai/agents/geo_agents.py

@@ -190,6 +190,8 @@ class GeoAgent(Agent):
         *,
         model: str = "llama3.1",
         map_instance: Optional[leafmap.Map] = None,
+        system_prompt: str = "default",
+        model_args: dict = None,
         **kwargs: Any,
     ) -> None:
         """Initialize the GeoAgent.
@@ -197,21 +199,25 @@ class GeoAgent(Agent):
         Args:
             model: Model identifier (default: "llama3.1").
             map_instance: Optional existing map instance.
+            model_args: Additional keyword arguments for the model.
             **kwargs: Additional keyword arguments for the model.
         """
         self.session: MapSession = MapSession(map_instance)
         self.tools: MapTools = MapTools(self.session)
 
+        if model_args is None:
+            model_args = {}
+
         # --- save a model factory we can call each turn ---
         if model == "llama3.1":
             self._model_factory: Callable[[], OllamaModel] = (
                 lambda: create_ollama_model(
-                    host="http://localhost:11434", model_id=model, **kwargs
+                    host="http://localhost:11434", model_id=model, **model_args
                 )
             )
         elif isinstance(model, str):
             self._model_factory: Callable[[], BedrockModel] = (
-                lambda: create_bedrock_model(model_id=model, **kwargs)
+                lambda: create_bedrock_model(model_id=model, **model_args)
             )
         elif isinstance(model, OllamaModel):
             # Extract configuration from existing OllamaModel and create new instances
@@ -220,7 +226,7 @@ class GeoAgent(Agent):
             client_args = model.client_args
             self._model_factory: Callable[[], OllamaModel] = (
                 lambda: create_ollama_model(
-                    host=host, model_id=model_id, client_args=client_args, **kwargs
+                    host=host, model_id=model_id, client_args=client_args, **model_args
                 )
             )
         elif isinstance(model, OpenAIModel):
@@ -229,7 +235,7 @@ class GeoAgent(Agent):
             client_args = model.client_args.copy()
             self._model_factory: Callable[[], OpenAIModel] = (
                 lambda mid=model_id, client_args=client_args: create_openai_model(
-                    model_id=mid, client_args=client_args, **kwargs
+                    model_id=mid, client_args=client_args, **model_args
                 )
             )
         elif isinstance(model, AnthropicModel):
@@ -238,7 +244,7 @@ class GeoAgent(Agent):
             client_args = model.client_args.copy()
             self._model_factory: Callable[[], AnthropicModel] = (
                 lambda mid=model_id, client_args=client_args: create_anthropic_model(
-                    model_id=mid, client_args=client_args, **kwargs
+                    model_id=mid, client_args=client_args, **model_args
                 )
             )
         else:
@@ -247,6 +253,28 @@ class GeoAgent(Agent):
         # build initial model (first turn)
         model = self._model_factory()
 
+        if system_prompt == "default":
+            system_prompt = """
+            You are a map control agent. Call tools with MINIMAL parameters only.
+
+            CRITICAL: Treat all kwargs parameters as optional parameters.
+            CRITICAL: NEVER include optional parameters unless user explicitly asks for them.
+
+            TOOL CALL RULES:
+            - zoom_to(zoom=N) - ONLY zoom parameter, OMIT options completely
+            - add_cog_layer(url='X') - NEVER include bands, nodata, opacity, etc.
+            - fly_to(longitude=N, latitude=N) - NEVER include zoom parameter
+            - add_basemap(name='X') - NEVER include any other parameters
+            - add_marker(lng_lat=[lon,lat]) - NEVER include popup or options
+
+            - remove_layer(name='X') - call get_layer_names() to get the layer name closest to
+            the name of the layer you want to remove before calling this tool
+
+            - add_overture_3d_buildings(kwargs={}) - kwargs parameter required by tool validation
+            FORBIDDEN: Optional parameters, string representations like '{}' or '[1,2,3]'
+            REQUIRED: Minimal tool calls with only what's absolutely necessary
+            """
+
         super().__init__(
             name="Leafmap Visualization Agent",
             model=model,
@@ -276,20 +304,7 @@ class GeoAgent(Agent):
                 self.tools.add_marker,
                 self.tools.set_pitch,
             ],
-            system_prompt="You are a map control agent. Call tools with MINIMAL parameters only.\n\n"
-            + "CRITICAL: Treat all kwargs parameters as optional parameters.\n"
-            + "CRITICAL: NEVER include optional parameters unless user explicitly asks for them.\n\n"
-            + "TOOL CALL RULES:\n"
-            + "- zoom_to(zoom=N) - ONLY zoom parameter, OMIT options completely\n"
-            + "- add_cog_layer(url='X') - NEVER include bands, nodata, opacity, etc.\n"
-            + "- fly_to(longitude=N, latitude=N) - NEVER include zoom parameter\n"
-            + "- add_basemap(name='X') - NEVER include any other parameters\n"
-            + "- add_marker(lng_lat=[lon,lat]) - NEVER include popup or options\n\n"
-            + "- remove_layer(name='X') - call get_layer_names() to get the layer name closest to"
-            + "the name of the layer you want to remove before calling this tool\n\n"
-            + "- add_overture_3d_buildings(kwargs={}) - kwargs parameter required by tool validation\n"
-            + "FORBIDDEN: Optional parameters, string representations like '{}' or '[1,2,3]'\n"
-            + "REQUIRED: Minimal tool calls with only what's absolutely necessary",
+            system_prompt=system_prompt,
             callback_handler=None,
         )
 
@@ -389,7 +404,7 @@ class GeoAgent(Agent):
                 ),
                 (
                     "Add GeoJSON",
-                    "Add vector layer: https://github.com/opengeos/datasets/releases/download/us/us_states.geojson",
+                    "Add GeoJSON layer: https://github.com/opengeos/datasets/releases/download/us/us_states.geojson",
                 ),
                 ("Remove layer", "Remove layer OpenTopoMap"),
                 ("Save map", "Save the map as demo.html and return the path"),

geoai/agents/map_tools.py

@@ -115,7 +115,7 @@ class MapTools:
         visible: bool = True,
         bands: Optional[List[int]] = None,
         nodata: Optional[Union[int, float]] = 0,
-        titiler_endpoint: str = "https://giswqs-titiler-endpoint.hf.space",
+        titiler_endpoint: str = None,
     ) -> str:
         """Add a Cloud Optimized GeoTIFF (COG) layer to the map.
 

geoai/train.py

@@ -2625,6 +2625,8 @@ def semantic_inference_on_geotiff(
     num_channels: int = 3,
     num_classes: int = 2,
     device: Optional[torch.device] = None,
+    probability_path: Optional[str] = None,
+    probability_threshold: Optional[float] = None,
     quiet: bool = False,
     **kwargs: Any,
 ) -> Tuple[str, float]:
@@ -2641,6 +2643,11 @@ def semantic_inference_on_geotiff(
         num_channels (int): Number of channels to use from the input image.
         num_classes (int): Number of classes in the model output.
         device (torch.device, optional): Device to run inference on.
+        probability_path (str, optional): Path to save probability map. If provided,
+            the normalized class probabilities will be saved as a multi-band raster.
+        probability_threshold (float, optional): Probability threshold for binary classification.
+            Only used when num_classes=2. If provided, pixels with class 1 probability >= threshold
+            are classified as class 1, otherwise class 0. If None (default), uses argmax.
         quiet (bool): If True, suppress progress bar. Defaults to False.
         **kwargs: Additional arguments.
 
@@ -2811,10 +2818,19 @@ def semantic_inference_on_geotiff(
                     / count_accumulator[valid_pixels]
                 )
 
-            # Take argmax to get final class predictions
-            mask[valid_pixels] = np.argmax(
-                normalized_probs[:, valid_pixels], axis=0
-            ).astype(np.uint8)
+            # Apply threshold for binary classification or use argmax
+            if probability_threshold is not None and num_classes == 2:
+                # Use threshold: classify as class 1 if probability >= threshold
+                mask[valid_pixels] = (
+                    normalized_probs[1, valid_pixels] >= probability_threshold
+                ).astype(np.uint8)
+                if not quiet:
+                    print(f"Using probability threshold: {probability_threshold}")
+            else:
+                # Take argmax to get final class predictions
+                mask[valid_pixels] = np.argmax(
+                    normalized_probs[:, valid_pixels], axis=0
+                ).astype(np.uint8)
 
             # Check class distribution in predictions (summary only)
             unique_classes, class_counts = np.unique(
@@ -2839,6 +2855,29 @@ def semantic_inference_on_geotiff(
         if not quiet:
             print(f"Saved prediction to {output_path}")
 
+        # Save probability map if requested
+        if probability_path is not None:
+            prob_dir = os.path.abspath(os.path.dirname(probability_path))
+            os.makedirs(prob_dir, exist_ok=True)
+
+            # Prepare probability output metadata
+            prob_meta = meta.copy()
+            prob_meta.update({"count": num_classes, "dtype": "float32"})
+
+            # Save normalized probabilities
+            with rasterio.open(probability_path, "w", **prob_meta) as dst:
+                for class_idx in range(num_classes):
+                    # Normalize probabilities
+                    prob_band = np.zeros((height, width), dtype=np.float32)
+                    prob_band[valid_pixels] = (
+                        prob_accumulator[class_idx, valid_pixels]
+                        / count_accumulator[valid_pixels]
+                    )
+                    dst.write(prob_band, class_idx + 1)
+
+            if not quiet:
+                print(f"Saved probability map to {probability_path}")
+
         return output_path, inference_time
 
 
@@ -2853,6 +2892,8 @@ def semantic_inference_on_image(
     num_classes: int = 2,
     device: Optional[torch.device] = None,
     binary_output: bool = True,
+    probability_path: Optional[str] = None,
+    probability_threshold: Optional[float] = None,
     quiet: bool = False,
     **kwargs: Any,
 ) -> Tuple[str, float]:
@@ -2870,6 +2911,11 @@ def semantic_inference_on_image(
         num_classes (int): Number of classes in the model output.
         device (torch.device, optional): Device to run inference on.
         binary_output (bool): If True, convert multi-class output to binary (class > 0).
+        probability_path (str, optional): Path to save probability map. If provided,
+            the normalized class probabilities will be saved as a multi-band raster.
+        probability_threshold (float, optional): Probability threshold for binary classification.
+            Only used when num_classes=2. If provided, pixels with class 1 probability >= threshold
+            are classified as class 1, otherwise class 0. If None (default), uses argmax.
         quiet (bool): If True, suppress progress bar. Defaults to False.
         **kwargs: Additional arguments.
 
@@ -3056,10 +3102,19 @@ def semantic_inference_on_image(
                     / count_accumulator[valid_pixels]
                 )
 
-            # Take argmax to get final class predictions
-            mask[valid_pixels] = np.argmax(
-                normalized_probs[:, valid_pixels], axis=0
-            ).astype(np.uint8)
+            # Apply threshold for binary classification or use argmax
+            if probability_threshold is not None and num_classes == 2:
+                # Use threshold: classify as class 1 if probability >= threshold
+                mask[valid_pixels] = (
+                    normalized_probs[1, valid_pixels] >= probability_threshold
+                ).astype(np.uint8)
+                if not quiet:
+                    print(f"Using probability threshold: {probability_threshold}")
+            else:
+                # Take argmax to get final class predictions
+                mask[valid_pixels] = np.argmax(
+                    normalized_probs[:, valid_pixels], axis=0
+                ).astype(np.uint8)
 
             # Check class distribution in predictions before binary conversion
             unique_classes, class_counts = np.unique(mask, return_counts=True)
@@ -3116,6 +3171,40 @@ def semantic_inference_on_image(
             if not quiet:
                 print(f"Saved prediction to {output_path}")
 
+        # Save probability map if requested
+        if probability_path is not None:
+            prob_dir = os.path.abspath(os.path.dirname(probability_path))
+            os.makedirs(prob_dir, exist_ok=True)
+
+            # For regular images, we'll save as a multi-channel TIFF
+            # since we need to preserve floating point values
+            import rasterio
+            from rasterio.transform import from_bounds
+
+            # Create a simple affine transform (identity transform for pixel coordinates)
+            transform = from_bounds(0, 0, width, height, width, height)
+
+            # Prepare probability output metadata
+            prob_meta = {
+                "driver": "GTiff",
+                "height": height,
+                "width": width,
+                "count": num_classes,
+                "dtype": "float32",
+                "transform": transform,
+            }
+
+            # Save normalized probabilities
+            with rasterio.open(probability_path, "w", **prob_meta) as dst:
+                for class_idx in range(num_classes):
+                    # Normalize probabilities
+                    prob_band = np.zeros((height, width), dtype=np.float32)
+                    prob_band[valid_pixels] = normalized_probs[class_idx, valid_pixels]
+                    dst.write(prob_band, class_idx + 1)
+
+            if not quiet:
+                print(f"Saved probability map to {probability_path}")
+
         return output_path, inference_time
 
 
@@ -3131,6 +3220,8 @@ def semantic_segmentation(
     overlap: int = 256,
     batch_size: int = 4,
     device: Optional[torch.device] = None,
+    probability_path: Optional[str] = None,
+    probability_threshold: Optional[float] = None,
     quiet: bool = False,
     **kwargs: Any,
 ) -> None:
@@ -3152,6 +3243,12 @@ def semantic_segmentation(
         overlap (int): Overlap between adjacent windows.
         batch_size (int): Batch size for inference.
         device (torch.device, optional): Device to run inference on.
+        probability_path (str, optional): Path to save probability map. If provided,
+            the normalized class probabilities will be saved as a multi-band raster.
+        probability_threshold (float, optional): Probability threshold for binary classification.
+            Only used when num_classes=2. If provided, pixels with class 1 probability >= threshold
+            are classified as class 1, otherwise class 0. If None (default), uses argmax.
+            Must be between 0 and 1.
         quiet (bool): If True, suppress progress bar. Defaults to False.
         **kwargs: Additional arguments.
 
@@ -3205,6 +3302,15 @@ def semantic_segmentation(
     model.to(device)
     model.eval()
 
+    # Validate probability_threshold
+    if probability_threshold is not None:
+        if not (0 <= probability_threshold <= 1):
+            raise ValueError("probability_threshold must be between 0 and 1")
+        if num_classes != 2:
+            raise ValueError(
+                "probability_threshold is only supported for binary classification (num_classes=2)"
+            )
+
     # Use appropriate inference function based on file format
     if is_geotiff:
         semantic_inference_on_geotiff(
@@ -3217,6 +3323,8 @@ def semantic_segmentation(
             num_channels=num_channels,
             num_classes=num_classes,
             device=device,
+            probability_path=probability_path,
+            probability_threshold=probability_threshold,
             quiet=quiet,
             **kwargs,
         )
@@ -3235,6 +3343,8 @@ def semantic_segmentation(
             num_classes=num_classes,
             device=device,
             binary_output=True,  # Convert to binary output for better visualization
+            probability_path=probability_path,
+            probability_threshold=probability_threshold,
             quiet=quiet,
             **kwargs,
         )

geoai/utils.py

@@ -7522,7 +7522,11 @@ def write_colormap(
 
 
 def plot_performance_metrics(
-    history_path: str, figsize: Tuple[int, int] = (15, 5), verbose: bool = True
+    history_path: str,
+    figsize: Tuple[int, int] = (15, 5),
+    verbose: bool = True,
+    save_path: Optional[str] = None,
+    kwargs: Optional[Dict] = None,
 ) -> None:
     """Plot performance metrics from a history object.
 
@@ -7531,6 +7535,8 @@ def plot_performance_metrics(
         figsize: The figure size.
         verbose: Whether to print the best and final metrics.
     """
+    if kwargs is None:
+        kwargs = {}
     history = torch.load(history_path)
 
     # Handle different key naming conventions
@@ -7579,6 +7585,14 @@ def plot_performance_metrics(
         plt.grid(True)
 
     plt.tight_layout()
+
+    if save_path:
+        if "dpi" not in kwargs:
+            kwargs["dpi"] = 150
+        if "bbox_inches" not in kwargs:
+            kwargs["bbox_inches"] = "tight"
+        plt.savefig(save_path, **kwargs)
+
     plt.show()
 
     if verbose:

{geoai_py-0.13.0.dist-info → geoai_py-0.13.2.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: geoai-py
-Version: 0.13.0
+Version: 0.13.2
 Summary: A Python package for using Artificial Intelligence (AI) with geospatial data
 Author-email: Qiusheng Wu <giswqs@gmail.com>
 License: MIT License

{geoai_py-0.13.0.dist-info → geoai_py-0.13.2.dist-info}/RECORD

@@ -1,4 +1,4 @@
-geoai/__init__.py,sha256=rKb_wHNQVHEFMzIYMkEXwp3_SlYN2YBzPvPFunMlZbs,3851
+geoai/__init__.py,sha256=4gwrJIsRBRP9mcJZMjX3TXxCtV_61kmf7ewRRkcDjiM,3851
 geoai/change_detection.py,sha256=XkJjMEU1nD8uX3-nQy7NEmz8cukVeSaRxKJHlrv8xPM,59636
 geoai/classify.py,sha256=0DcComVR6vKU4qWtH2oHVeXc7ZTcV0mFvdXRtlNmolo,35637
 geoai/detectron2.py,sha256=dOOFM9M9-6PV8q2A4-mnIPrz7yTo-MpEvDiAW34nl0w,14610
@@ -11,14 +11,14 @@ geoai/map_widgets.py,sha256=QLmkILsztNaRXRULHKOd7Glb7S0pEWXSK9-P8S5AuzQ,5856
 geoai/sam.py,sha256=O6S-kGiFn7YEcFbfWFItZZQOhnsm6-GlunxQLY0daEs,34345
 geoai/segment.py,sha256=yBGTxA-ti8lBpk7WVaBOp6yP23HkaulKJQk88acrmZ0,43788
 geoai/segmentation.py,sha256=7yEzBSKCyHW1dNssoK0rdvhxi2IXsIQIFSga817KdI4,11535
-geoai/train.py,sha256=r9eioaBpc2eg6hckkGVI3aGhQZffKas_UVRj-AWruu8,136049
-geoai/utils.py,sha256=ECl5IWkGQA7-_50Ws27T5EMIg_88EJuIktI8gprviBw,300932
+geoai/train.py,sha256=QDG0z6VmvHmzJC-oujZb2YSxnNEBloU_lmEgSygPM6U,141818
+geoai/utils.py,sha256=lpyhytBeDLiqWz31syeRvpbT5AUn3cOblKU57uDD9sU,301265
 geoai/agents/__init__.py,sha256=NndUtQ5-i8Zuim8CJftCZYKbCvrkDXj9iLVtiBtc_qE,178
-geoai/agents/geo_agents.py,sha256=6vr5Asp18q_JD4aG9ZmebS7eNEseGsfrlZ0ChRKjLXk,21177
-geoai/agents/map_tools.py,sha256=7aepLzzKHYPEyj43lTTgWZJme8DQP6dgPUNK4yeRqNE,60707
-geoai_py-0.13.0.dist-info/licenses/LICENSE,sha256=vN2L5U7cZ6ZkOHFmc8WiGlsogWsZc5dllMeNxnKVOZg,1070
-geoai_py-0.13.0.dist-info/METADATA,sha256=FP9x9Q2HS08x__vjOtpLlHNX2Q7NGGe7bnRCN4EMCOw,10345
-geoai_py-0.13.0.dist-info/WHEEL,sha256=JNWh1Fm1UdwIQV075glCn4MVuCRs0sotJIq-J6rbxCU,109
-geoai_py-0.13.0.dist-info/entry_points.txt,sha256=uGp3Az3HURIsRHP9v-ys0hIbUuBBNUfXv6VbYHIXeg4,41
-geoai_py-0.13.0.dist-info/top_level.txt,sha256=1YkCUWu-ii-0qIex7kbwAvfei-gos9ycyDyUCJPNWHY,6
-geoai_py-0.13.0.dist-info/RECORD,,
+geoai/agents/geo_agents.py,sha256=4tLntKBL_FgTQsUVzReP9acbYotnfjMRc5BYwW9WEyE,21431
+geoai/agents/map_tools.py,sha256=OK5uB0VUHjjUnc-DYRy2CQ__kyUIARSCPBucGabO0Xw,60669
+geoai_py-0.13.2.dist-info/licenses/LICENSE,sha256=vN2L5U7cZ6ZkOHFmc8WiGlsogWsZc5dllMeNxnKVOZg,1070
+geoai_py-0.13.2.dist-info/METADATA,sha256=wGO4OHdq1ubtroaqbEciKq1jHmJD8C_krbZTJByFP98,10345
+geoai_py-0.13.2.dist-info/WHEEL,sha256=JNWh1Fm1UdwIQV075glCn4MVuCRs0sotJIq-J6rbxCU,109
+geoai_py-0.13.2.dist-info/entry_points.txt,sha256=uGp3Az3HURIsRHP9v-ys0hIbUuBBNUfXv6VbYHIXeg4,41
+geoai_py-0.13.2.dist-info/top_level.txt,sha256=1YkCUWu-ii-0qIex7kbwAvfei-gos9ycyDyUCJPNWHY,6
+geoai_py-0.13.2.dist-info/RECORD,,