diff --git a/pyproject.toml b/pyproject.toml
index 8d88732..b0473ff 100644
--- a/pyproject.toml
+++ b/pyproject.toml
@@ -38,7 +38,8 @@ dependencies = [
     "scipy >= 1.10.0",
     "tqdm >= 4.65.0",
     "importlib_resources >= 6.0.0",
-    "pyarrow >= 11.0.0"
+    "pyarrow >= 14.0.1",
+    "humanize >= 4.9.0"
 ]
 
 [project.urls]
diff --git a/tests/minimal_requirements.txt b/tests/minimal_requirements.txt
index 7d35850..3838010 100644
--- a/tests/minimal_requirements.txt
+++ b/tests/minimal_requirements.txt
@@ -3,4 +3,4 @@ pandas==2.0.0
 scipy==1.10.0
 tqdm==4.65.0
 importlib_resources==6.0.0
-pyarrow==11.0.0
+pyarrow==14.0.1
diff --git a/tests/test_metadata.py b/tests/test_metadata.py
index 9321ee4..8c19dcc 100644
--- a/tests/test_metadata.py
+++ b/tests/test_metadata.py
@@ -238,6 +238,40 @@ def test_add_relationships_new_table(multitable_metadata):
         )
 
 
+def test_sample_data_single():
+    dataframe, ml_types, _, primary_key, time_index = generate_mock_data(
+        tables=["products"],
+    )
+    dataframe = dataframe["products"]
+    primary_key = "key does not exist"
+    match = "does not exist in sample data's columns"
+    with pytest.raises(ValueError, match=match):
+        SingleTableMetadata(
+            ml_types=ml_types,
+            primary_key=primary_key,
+            time_index=time_index,
+            sample_data=dataframe,
+        )
+
+
+def test_sample_data_multi():
+    dataframes, ml_types, relationships, primary_keys, time_indices = (
+        generate_mock_data(
+            tables=["products", "logs"],
+        )
+    )
+    primary_keys["products"] = "key does not exist"
+    match = "does not exist in sample data's table"
+    with pytest.raises(ValueError, match=match):
+        MultiTableMetadata(
+            ml_types=ml_types,
+            primary_keys=primary_keys,
+            time_indices=time_indices,
+            relationships=relationships,
+            sample_data=dataframes,
+        )
+
+
 def verify_ml_types(metadata, expected_ml_types):
     if metadata.get_metadata_type() == "single":
         for key in expected_ml_types:
diff --git a/trane/core/problem.py b/trane/core/problem.py
index dff7d07..da40570 100644
--- a/trane/core/problem.py
+++ b/trane/core/problem.py
@@ -1,3 +1,4 @@
+import humanize
 import pandas as pd
 
 from trane.core.utils import calculate_target_values
@@ -188,8 +189,10 @@ def __str__(self):
         description += filter_op.generate_description()
 
         if self.window_size:
-            description += " " + "in next {} days".format(
-                self.window_size,
+            window_size = pd.to_timedelta(self.window_size)
+            human_readble = humanize.naturaldelta(window_size)
+            description += " " + "in next {}".format(
+                human_readble,
             )
         return description
 
diff --git a/trane/metadata/__init__.py b/trane/metadata/__init__.py
index fe335db..b2a0ae3 100644
--- a/trane/metadata/__init__.py
+++ b/trane/metadata/__init__.py
@@ -1 +1 @@
-from trane.metadata.metadata import *
+from trane.metadata.metadata import SingleTableMetadata, MultiTableMetadata
diff --git a/trane/metadata/metadata.py b/trane/metadata/metadata.py
index 1c924d4..1a29950 100644
--- a/trane/metadata/metadata.py
+++ b/trane/metadata/metadata.py
@@ -33,8 +33,10 @@ def __init__(
         primary_key: str = None,
         time_index: str = None,
         original_multi_table_metadata=None,
+        sample_data: pd.DataFrame = None,
     ):
         self.ml_types = _parse_ml_types(ml_types, type_=self.get_metadata_type())
+        self.sample_data = sample_data
         self.primary_key = None
         if primary_key:
             self.set_primary_key(primary_key)
@@ -53,7 +55,14 @@ def __repr__(self):
         return result
 
     def set_primary_key(self, primary_key):
-        if primary_key not in self.ml_types:
+        if (
+            self.sample_data is not None
+            and primary_key not in self.sample_data.columns.tolist()
+        ):
+            raise ValueError(
+                f"Index {primary_key} does not exist in sample data's columns",
+            )
+        elif primary_key not in self.ml_types:
             raise ValueError("Index does not exist in ml_types")
         elif (
             self.primary_key
@@ -109,11 +118,13 @@ class MultiTableMetadata(BaseMetadata):
     def __init__(
         self,
         ml_types: dict = None,
-        primary_keys=None,
-        time_indices=None,
+        primary_keys: dict = None,
+        time_indices: dict = None,
         relationships: list = None,
+        sample_data: pd.DataFrame = None,
     ):
         self.ml_types = _parse_ml_types(ml_types, type_=self.get_metadata_type())
+        self.sample_data = sample_data
         self.primary_keys = {}
         if primary_keys:
             self.set_primary_keys(primary_keys)
@@ -200,7 +211,14 @@ def check_if_table_exists(self, table):
             raise ValueError(f"Table: {table} does not exist")
 
     def check_if_column_exists(self, table, column):
-        if column not in self.ml_types[table]:
+        if (
+            self.sample_data is not None
+            and column not in self.sample_data[table].columns
+        ):
+            raise ValueError(
+                f"Column: {column} does not exist in sample data's table: {table}",
+            )
+        elif column not in self.ml_types[table]:
             raise ValueError(f"Column: {column} does not exist in Table: {table}")
 
     def remove_table(self, table):