mostlyai-engine 2.0.0__tar.gz → 2.0.1__tar.gz

{mostlyai_engine-2.0.0 → mostlyai_engine-2.0.1}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: mostlyai-engine
-Version: 2.0.0
+Version: 2.0.1
 Summary: Synthetic Data Engine
 Project-URL: homepage, https://github.com/mostly-ai/mostlyai-engine
 Project-URL: repository, https://github.com/mostly-ai/mostlyai-engine
@@ -173,8 +173,12 @@ data_with_missings.loc[0:299, "age"] = pd.NA
 data_with_missings.loc[0:199, "race"] = pd.NA
 data_with_missings.loc[100:299, "income"] = pd.NA
 
-# impute missing values
-argn.impute(data_with_missings)
+# impute missing values each with a random sample
+data_imputed = argn.impute(data_with_missings)
+
+# impute missing values each with their point estimates
+data_imputed = argn.impute(data_with_missings, n_draws=100)
+
 ```
 
 ### Predictions / Classification
@@ -185,10 +189,10 @@ Predict any categorical target column:
 from sklearn.metrics import accuracy_score, roc_auc_score
 
 # predict class labels for a categorical
-predictions = argn.predict(data_test, target="income", n_draws=10, agg_fn="mode")
+predictions = argn.predict(data_test, target="income", n_draws=100, agg_fn="mode")
 
 # predict class probabilities for a categorical
-probabilities = argn.predict_proba(data_test, target="income", n_draws=10)
+probabilities = argn.predict_proba(data_test, target="income", n_draws=100)
 
 # evaluate performance
 accuracy = accuracy_score(data_test["income"], predictions)

{mostlyai_engine-2.0.0 → mostlyai_engine-2.0.1}/README.md

@@ -122,8 +122,12 @@ data_with_missings.loc[0:299, "age"] = pd.NA
 data_with_missings.loc[0:199, "race"] = pd.NA
 data_with_missings.loc[100:299, "income"] = pd.NA
 
-# impute missing values
-argn.impute(data_with_missings)
+# impute missing values each with a random sample
+data_imputed = argn.impute(data_with_missings)
+
+# impute missing values each with their point estimates
+data_imputed = argn.impute(data_with_missings, n_draws=100)
+
 ```
 
 ### Predictions / Classification
@@ -134,10 +138,10 @@ Predict any categorical target column:
 from sklearn.metrics import accuracy_score, roc_auc_score
 
 # predict class labels for a categorical
-predictions = argn.predict(data_test, target="income", n_draws=10, agg_fn="mode")
+predictions = argn.predict(data_test, target="income", n_draws=100, agg_fn="mode")
 
 # predict class probabilities for a categorical
-probabilities = argn.predict_proba(data_test, target="income", n_draws=10)
+probabilities = argn.predict_proba(data_test, target="income", n_draws=100)
 
 # evaluate performance
 accuracy = accuracy_score(data_test["income"], predictions)

{mostlyai_engine-2.0.0 → mostlyai_engine-2.0.1}/mostlyai/engine/__init__.py

@@ -34,7 +34,7 @@ __all__ = [
     "TabularARGN",
     "LanguageModel",
 ]
-__version__ = "2.0.0"
+__version__ = "2.0.1"
 
 # suppress specific warning related to os.fork() in multi-threaded processes
 warnings.filterwarnings("ignore", category=DeprecationWarning, message=".*multi-threaded.*fork.*")

{mostlyai_engine-2.0.0 → mostlyai_engine-2.0.1}/mostlyai/engine/_encoding_types/tabular/numeric.py

@@ -105,6 +105,19 @@ def _type_safe_numeric_series(numeric_array: np.ndarray | list, pd_dtype: str) -
     return pd.Series(np.array([v for v in numeric_array]).astype(np_dtype), dtype=pd_dtype)
 
 
+def _cast_based_on_min_decimal(values: pd.Series, min_decimal: int) -> pd.Series:
+    # try to convert to int when min_decimal is 0, if possible
+    dtype = "Float64" if min_decimal < 0 else "Int64"
+    if dtype == "Int64":
+        values = values.round()
+    try:
+        values = values.astype(dtype)
+    except TypeError:
+        if dtype == "Int64":
+            values = values.astype("Float64")  # if couldn't safely convert to int, stick to float
+    return values
+
+
 def split_sub_columns_digit(
     values: pd.Series,
     max_decimal=NUMERIC_DIGIT_MAX_DECIMAL,
@@ -295,6 +308,9 @@ def analyze_reduce_numeric(
             encoding_type = ModelEncodingType.tabular_numeric_binned
 
     if encoding_type == ModelEncodingType.tabular_numeric_discrete:
+        if min_decimal >= 0:
+            # remove decimal part from categories
+            categories = [str(cat).split(".")[0] for cat in categories]
         # add NULL token if NaN values exist
         if has_nan:
             categories = [NUMERIC_DISCRETE_NULL_TOKEN] + categories
@@ -368,6 +384,8 @@ def analyze_reduce_numeric(
 
 
 def encode_numeric(values: pd.Series, stats: dict, _: pd.Series | None = None) -> pd.DataFrame:
+    values = safe_convert_numeric(values)
+
     if stats["encoding_type"] == ModelEncodingType.tabular_numeric_discrete:
         df = _encode_numeric_discrete(values, stats)
     elif stats["encoding_type"] == ModelEncodingType.tabular_numeric_digit:
@@ -380,31 +398,21 @@ def encode_numeric(values: pd.Series, stats: dict, _: pd.Series | None = None) -
 
 
 def _encode_numeric_discrete(values: pd.Series, stats: dict, _: pd.Series | None = None) -> pd.DataFrame:
-    values = safe_convert_numeric(values)
+    values = _cast_based_on_min_decimal(values, stats["min_decimal"])
     df = encode_categorical(values, stats)
     return df
 
 
 def _encode_numeric_digit(values: pd.Series, stats: dict, _: pd.Series | None = None) -> pd.DataFrame:
-    values = safe_convert_numeric(values)
-    # try to convert to int, if possible
-    dtype = "Int64" if stats["min_decimal"] == 0 else "Float64"
-    if dtype == "Int64":
-        values = values.round()
-    try:
-        values = values.astype(dtype)
-    except TypeError:
-        if dtype == "Int64":  # if couldn't safely convert to int, stick to float
-            dtype = "Float64"
-            values = values.astype(dtype)
+    values = _cast_based_on_min_decimal(values, stats["min_decimal"])
     # reset index, as `values.mask` can throw errors for misaligned indices
     values.reset_index(drop=True, inplace=True)
     # replace extreme values with min/max
     if stats["min"] is not None:
-        reduced_min = _type_safe_numeric_series([stats["min"]], dtype).iloc[0]
+        reduced_min = _type_safe_numeric_series([stats["min"]], values.dtype).iloc[0]
         values = values.where((values.isna()) | (values >= reduced_min), reduced_min)
     if stats["max"] is not None:
-        reduced_max = _type_safe_numeric_series([stats["max"]], dtype).iloc[0]
+        reduced_max = _type_safe_numeric_series([stats["max"]], values.dtype).iloc[0]
         values = values.where((values.isna()) | (values <= reduced_max), reduced_max)
     values, nan_mask = impute_from_non_nan_distribution(values, stats)
     # split to sub_columns
@@ -431,7 +439,6 @@ def _encode_numeric_digit(values: pd.Series, stats: dict, _: pd.Series | None =
 
 
 def _encode_numeric_binned(values: pd.Series, stats: dict, _: pd.Series | None = None) -> pd.DataFrame:
-    values = safe_convert_numeric(values)
     bins = stats["bins"].copy()
     min_value = bins[0]
     max_value = bins[-1]

{mostlyai_engine-2.0.0 → mostlyai_engine-2.0.1}/mostlyai/engine/_tabular/encoding.py

@@ -13,6 +13,7 @@
 # limitations under the License.
 
 import logging
+import os
 import time
 from pathlib import Path
 
@@ -233,6 +234,7 @@ def encode_df(
                 values=df[column],
                 column_stats=column_stats,
                 context_keys=context_keys,
+                parent_pid=os.getpid(),
             )
         )
     if delayed_encodes:
@@ -247,8 +249,10 @@ def _encode_col(
     values: pd.Series,
     column_stats: dict,
     context_keys: pd.Series | None = None,
+    parent_pid: int | None = None,
 ) -> pd.DataFrame:
-    set_random_state(worker=True)
+    if os.getpid() != parent_pid:
+        set_random_state(worker=True)
     is_sequential_column = is_sequential(values)
     if is_sequential_column:
         # explode nested columns and encode the same way as flat columns

{mostlyai_engine-2.0.0 → mostlyai_engine-2.0.1}/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "mostlyai-engine"
-version = "2.0.0"
+version = "2.0.1"
 description = "Synthetic Data Engine"
 authors = [{ name = "MOSTLY AI", email = "dev@mostly.ai" }]
 requires-python = ">=3.10"