Merge pull request #173 from aimclub/fix/land_use_classification

Fix/land use classification

Author: vasilstar97

README.md

@@ -75,9 +75,10 @@ pip install blocksnet
 
 There are various extras to install:
 
-- `ml` - machine learning related packages (`torch`, `catboost`, etc).
+- `ml` - machine learning related packages (`catboost`, `lightgmb` etc).
+- `torch` - torch related packages (`torch`, `torch_geometric`).
 - `opt` - optimization problem related packages (`optuna`, `pymoo`).
-- `full` - all packages required by blocksnet methods (`ml` + `opt` extras).
+- `full` - all packages required by blocksnet methods (`ml` + `torch` + `opt` extras).
 - `ipynb` - jupyter notebook visualization packages (`matplotlib`, etc).
 - `tests` (DEVELOPMENT ONLY) - pytest related packages.
 - `docs` (DEVELOPMENT ONLY) - sphinx documentation related packages.
@@ -162,7 +163,7 @@ The repository includes the following directories and modules:
   - [`blocks`](blocksnet/blocks) - methods to generate, aggregate and process urban blocks.
   - [`config`](blocksnet/config) - config to handle logging and meta information (service types, land use relations, etc).
   - [`enums`](blocksnet/enums) - enums used within the code base.
-  - [`machine_learning`](blocksnet/machine_learning) - machine learning strategies for different packages. Requires `ml` extra (`pip install blocksnet[ml]`).
+  - [`machine_learning`](blocksnet/machine_learning) - machine learning strategies for different packages. Requires `ml` and `torch` (optional) extras (`pip install blocksnet[ml,torch]`).
   - [`optimization`](blocksnet/optimization) - optimization related methods. Requires the `opt` extra (`pip install blocksnet[opt]`).
   - [`preprocessing`](blocksnet/preprocessing) - data imputing and preprocessing.
   - [`relations`](blocksnet/relations) - methods to handle network graphs and accessibility matrices.

blocksnet/analysis/land_use/prediction/_strategy.py

@@ -50,5 +50,8 @@
     ("hgb", HistGradientBoostingClassifier(**MODEL_PARAMS["hgb"])),
 ]
 
-strategy = SKLearnVotingClassificationStrategy(estimators)
-strategy.load(ARTIFACTS_DIRECTORY)
+def get_default_strategy() -> SKLearnVotingClassificationStrategy:
+    strategy = SKLearnVotingClassificationStrategy(estimators)
+    strategy.load(ARTIFACTS_DIRECTORY)
+    print(ARTIFACTS_DIRECTORY)
+    return strategy

blocksnet/analysis/land_use/prediction/artifacts/meta.json

@@ -1 +1 @@
-{"strategy_cls": "SKLearnVotingClassificationStrategy", "model_cls": "VotingClassifier", "model_params": {"voting": "soft", "n_jobs": -1}}
+{"strategy_cls": "SKLearnVotingClassificationStrategy", "model_cls": "VotingClassifier", "model_params": {"voting": "soft", "n_jobs": -1}}

blocksnet/analysis/land_use/prediction/artifacts/model.joblib

@@ -1,22 +1,33 @@
 import pandas as pd
 from pandera.typing import Series
-from shapely import MultiPolygon, Polygon
+from shapely import Polygon
 from blocksnet.enums import LandUseCategory
 from blocksnet.utils.validation import GdfSchema, LandUseSchema
 
 
-class BlocksInputSchema(GdfSchema):
-    category: Series
+class BlocksRunSchema(GdfSchema):
+
+    # @classmethod
+    # def _after_validate(cls, df: pd.DataFrame) -> pd.DataFrame:
+    #     if not 'category' in df:
+    #         df['category'] = None
+    #     return df
 
     @classmethod
     def _geometry_types(cls):
-        return [Polygon | MultiPolygon]
+        return {Polygon}
+
+
+class BlocksTrainSchema(BlocksRunSchema):
+
+    category: Series
 
     @classmethod
     def _before_validate(cls, df: pd.DataFrame) -> pd.DataFrame:
         df = df.copy()
 
         if "category" in df.columns:
+
             def parse_category(c):
                 if isinstance(c, LandUseCategory):
                     return c
@@ -35,4 +46,3 @@ def to_category(lu):
             df["category"] = lu_df["land_use"].map(to_category)
 
         return df
-

blocksnet/analysis/land_use/prediction/core.py

@@ -46,22 +46,17 @@ version_scheme = "post-release"
 local_scheme = "no-local-version"
 
 [project.optional-dependencies]
-ml = [
-    "pygeoops",
-    "featuretools",
-    "torch",
-    "torch_geometric",
-    'catboost',
-    'xgboost',
-]
+ml = ["pygeoops", "featuretools", 'catboost', 'xgboost', 'lightgbm']
+torch = ["torch", "torch_geometric"]
 opt = ["optuna>=4.1.0,<5.0.0", "pymoo>=0.6.0,<1.0.0"]
-full = ["blocksnet[ml, opt]"]
+full = ["blocksnet[ml, torch, opt]"]
 
 ipynb = [
     "mapclassify==2.6.1",       #jupyter maps
     "matplotlib-inline==0.1.6", #jupyter plots
     "folium==0.14.0",           #jupyter maps
     "matplotlib>=3.9.0,<4.0.0",
+    "ipywidgets",
 ]
 
 tests = ["pytest==7.4.3", "pytest-cov==4.1.0"]