Merge pull request #2554 from plotly/trendline_fix

make trendlines more robust

Author: nicolaskruchten

Diff for: CHANGELOG.md

@@ -8,6 +8,7 @@ This project adheres to [Semantic Versioning](http://semver.org/).
 
 - Fixed special cases with `px.sunburst` and `px.treemap` with `path` input ([#2524](https://github.com/plotly/plotly.py/pull/2524))
 - Fixed bug in `hover_data` argument of `px` functions, when the column name is changed with labels and `hover_data` is a dictionary setting up a specific format for the hover data ([#2544](https://github.com/plotly/plotly.py/pull/2544)).
+- Made the Plotly Express `trendline` argument more robust and made it work with datetime `x` values ([#2554](https://github.com/plotly/plotly.py/pull/2554))
 
 ## [4.8.1] - 2020-05-28
 

Diff for: packages/python/plotly/plotly/express/_core.py

@@ -277,17 +277,35 @@ def make_trace_kwargs(args, trace_spec, trace_data, mapping_labels, sizeref):
                     attr_value in ["ols", "lowess"]
                     and args["x"]
                     and args["y"]
-                    and len(trace_data) > 1
+                    and len(trace_data[[args["x"], args["y"]]].dropna()) > 1
                 ):
                     import statsmodels.api as sm
 
                     # sorting is bad but trace_specs with "trendline" have no other attrs
                     sorted_trace_data = trace_data.sort_values(by=args["x"])
-                    y = sorted_trace_data[args["y"]]
-                    x = sorted_trace_data[args["x"]]
+                    y = sorted_trace_data[args["y"]].values
+                    x = sorted_trace_data[args["x"]].values
 
+                    x_is_date = False
                     if x.dtype.type == np.datetime64:
                         x = x.astype(int) / 10 ** 9  # convert to unix epoch seconds
+                        x_is_date = True
+                    elif x.dtype.type == np.object_:
+                        try:
+                            x = x.astype(np.float64)
+                        except ValueError:
+                            raise ValueError(
+                                "Could not convert value of 'x' ('%s') into a numeric type. "
+                                "If 'x' contains stringified dates, please convert to a datetime column."
+                                % args["x"]
+                            )
+                    if y.dtype.type == np.object_:
+                        try:
+                            y = y.astype(np.float64)
+                        except ValueError:
+                            raise ValueError(
+                                "Could not convert value of 'y' into a numeric type."
+                            )
 
                     if attr_value == "lowess":
                         # missing ='drop' is the default value for lowess but not for OLS (None)
@@ -298,25 +316,32 @@ def make_trace_kwargs(args, trace_spec, trace_data, mapping_labels, sizeref):
                         hover_header = "<b>LOWESS trendline</b><br><br>"
                     elif attr_value == "ols":
                         fit_results = sm.OLS(
-                            y.values, sm.add_constant(x.values), missing="drop"
+                            y, sm.add_constant(x), missing="drop"
                         ).fit()
                         trace_patch["y"] = fit_results.predict()
                         trace_patch["x"] = x[
                             np.logical_not(np.logical_or(np.isnan(y), np.isnan(x)))
                         ]
                         hover_header = "<b>OLS trendline</b><br>"
-                        hover_header += "%s = %g * %s + %g<br>" % (
-                            args["y"],
-                            fit_results.params[1],
-                            args["x"],
-                            fit_results.params[0],
-                        )
+                        if len(fit_results.params) == 2:
+                            hover_header += "%s = %g * %s + %g<br>" % (
+                                args["y"],
+                                fit_results.params[1],
+                                args["x"],
+                                fit_results.params[0],
+                            )
+                        else:
+                            hover_header += "%s = %g<br>" % (
+                                args["y"],
+                                fit_results.params[0],
+                            )
                         hover_header += (
                             "R<sup>2</sup>=%f<br><br>" % fit_results.rsquared
                         )
+                    if x_is_date:
+                        trace_patch["x"] = pd.to_datetime(trace_patch["x"] * 10 ** 9)
                     mapping_labels[get_label(args, args["x"])] = "%{x}"
                     mapping_labels[get_label(args, args["y"])] = "%{y} <b>(trend)</b>"
-
             elif attr_name.startswith("error"):
                 error_xy = attr_name[:7]
                 arr = "arrayminus" if attr_name.endswith("minus") else "array"

Diff for: packages/python/plotly/plotly/tests/test_core/test_px/test_trendline.py

@@ -1,14 +1,110 @@
 import plotly.express as px
 import numpy as np
+import pandas as pd
+import pytest
+from datetime import datetime
 
 
-def test_trendline_nan_values():
+@pytest.mark.parametrize("mode", ["ols", "lowess"])
+def test_trendline_results_passthrough(mode):
+    df = px.data.gapminder().query("continent == 'Oceania'")
+    fig = px.scatter(df, x="year", y="pop", color="country", trendline=mode)
+    assert len(fig.data) == 4
+    for trace in fig["data"][0::2]:
+        assert "trendline" not in trace.hovertemplate
+    for trendline in fig["data"][1::2]:
+        assert "trendline" in trendline.hovertemplate
+        if mode == "ols":
+            assert "R<sup>2</sup>" in trendline.hovertemplate
+    results = px.get_trendline_results(fig)
+    if mode == "ols":
+        assert len(results) == 2
+        assert results["country"].values[0] == "Australia"
+        assert results["country"].values[0] == "Australia"
+        au_result = results["px_fit_results"].values[0]
+        assert len(au_result.params) == 2
+    else:
+        assert len(results) == 0
+
+
+@pytest.mark.parametrize("mode", ["ols", "lowess"])
+def test_trendline_enough_values(mode):
+    fig = px.scatter(x=[0, 1], y=[0, 1], trendline=mode)
+    assert len(fig.data) == 2
+    assert len(fig.data[1].x) == 2
+    fig = px.scatter(x=[0], y=[0], trendline=mode)
+    assert len(fig.data) == 2
+    assert fig.data[1].x is None
+    fig = px.scatter(x=[0, 1], y=[0, None], trendline=mode)
+    assert len(fig.data) == 2
+    assert fig.data[1].x is None
+    fig = px.scatter(x=[0, 1], y=np.array([0, np.nan]), trendline=mode)
+    assert len(fig.data) == 2
+    assert fig.data[1].x is None
+    fig = px.scatter(x=[0, 1, None], y=[0, None, 1], trendline=mode)
+    assert len(fig.data) == 2
+    assert fig.data[1].x is None
+    fig = px.scatter(
+        x=np.array([0, 1, np.nan]), y=np.array([0, np.nan, 1]), trendline=mode
+    )
+    assert len(fig.data) == 2
+    assert fig.data[1].x is None
+    fig = px.scatter(x=[0, 1, None, 2], y=[1, None, 1, 2], trendline=mode)
+    assert len(fig.data) == 2
+    assert len(fig.data[1].x) == 2
+    fig = px.scatter(
+        x=np.array([0, 1, np.nan, 2]), y=np.array([1, np.nan, 1, 2]), trendline=mode
+    )
+    assert len(fig.data) == 2
+    assert len(fig.data[1].x) == 2
+
+
+@pytest.mark.parametrize("mode", ["ols", "lowess"])
+def test_trendline_nan_values(mode):
     df = px.data.gapminder().query("continent == 'Oceania'")
     start_date = 1970
     df["pop"][df["year"] < start_date] = np.nan
-    modes = ["ols", "lowess"]
-    for mode in modes:
-        fig = px.scatter(df, x="year", y="pop", color="country", trendline=mode)
-        for trendline in fig["data"][1::2]:
-            assert trendline.x[0] >= start_date
-            assert len(trendline.x) == len(trendline.y)
+    fig = px.scatter(df, x="year", y="pop", color="country", trendline=mode)
+    for trendline in fig["data"][1::2]:
+        assert trendline.x[0] >= start_date
+        assert len(trendline.x) == len(trendline.y)
+
+
+def test_no_slope_ols_trendline():
+    fig = px.scatter(x=[0, 1], y=[0, 1], trendline="ols")
+    assert "y = 1" in fig.data[1].hovertemplate  # then + x*(some small number)
+    results = px.get_trendline_results(fig)
+    params = results["px_fit_results"].iloc[0].params
+    assert np.all(np.isclose(params, [0, 1]))
+
+    fig = px.scatter(x=[1, 1], y=[0, 0], trendline="ols")
+    assert "y = 0" in fig.data[1].hovertemplate
+    results = px.get_trendline_results(fig)
+    params = results["px_fit_results"].iloc[0].params
+    assert np.all(np.isclose(params, [0]))
+
+    fig = px.scatter(x=[1, 2], y=[0, 0], trendline="ols")
+    assert "y = 0" in fig.data[1].hovertemplate
+    fig = px.scatter(x=[0, 0], y=[1, 1], trendline="ols")
+    assert "y = 0 * x + 1" in fig.data[1].hovertemplate
+    fig = px.scatter(x=[0, 0], y=[1, 2], trendline="ols")
+    assert "y = 0 * x + 1.5" in fig.data[1].hovertemplate
+
+
+@pytest.mark.parametrize("mode", ["ols", "lowess"])
+def test_trendline_on_timeseries(mode):
+    df = px.data.stocks()
+
+    with pytest.raises(ValueError) as err_msg:
+        px.scatter(df, x="date", y="GOOG", trendline=mode)
+    assert "Could not convert value of 'x' ('date') into a numeric type." in str(
+        err_msg.value
+    )
+
+    df["date"] = pd.to_datetime(df["date"])
+    fig = px.scatter(df, x="date", y="GOOG", trendline=mode)
+    assert len(fig.data) == 2
+    assert len(fig.data[0].x) == len(fig.data[1].x)
+    assert type(fig.data[0].x[0]) == datetime
+    assert type(fig.data[1].x[0]) == datetime
+    assert np.all(fig.data[0].x == fig.data[1].x)