fix(eda): fix bugs in log transformation

dataprep/eda/distribution/render.py

@@ -339,14 +339,31 @@ def hist_viz(
     )
 
     tooltips = [("Bin", "@intvl"), ("Frequency", "@freq"), ("Percent", "@pct{0.2f}%")]
-    fig = Figure(
-        plot_height=plot_height,
-        plot_width=plot_width,
-        title=col,
-        toolbar_location=None,
-        y_axis_type=yscale,
-    )
-    bottom = 0 if yscale == "linear" or df.empty else df["freq"].min() / 2
+    if yscale == "linear" or df.empty:
+        bottom = 0.0
+    elif yscale == "log" and df["freq"].min() == 0:  # freq >= 1 so we set 0.1 as lower bound
+        bottom = 0.1
+    else:
+        bottom = df["freq"].min() / 2
+
+    if yscale == "linear":
+        fig = Figure(
+            plot_height=plot_height,
+            plot_width=plot_width,
+            title=col,
+            toolbar_location=None,
+            y_axis_type=yscale,
+        )
+    else:
+        fig = Figure(
+            plot_height=plot_height,
+            plot_width=plot_width,
+            title=col,
+            toolbar_location=None,
+            y_axis_type=yscale,
+            y_range=(bottom, df["freq"].max()),
+        )
+
     fig.quad(
         source=df,
         left="left",

dataprep/tests/eda/test_plot.py

@@ -35,6 +35,7 @@ def simpledf() -> dd.DataFrame:
     df["e"] = pd.to_datetime(df["e"])
     # test when column is object but some cells are numerical
     df["g"] = pd.Series([0, "x"] * 500)
+    df["h"] = pd.Series(np.ones(1000))
 
     idx = np.arange(1000)
     np.random.shuffle(idx)
@@ -62,6 +63,7 @@ def test_sanity_compute_univariate(simpledf: dd.DataFrame) -> None:
 
 def test_sanity_compute_overview(simpledf: dd.DataFrame) -> None:
     plot(simpledf)
+    plot(simpledf, config={"hist.yscale": "log"})
 
 
 def test_sanity_compute_bivariate(simpledf: dd.DataFrame) -> None: