google-meridian 1.0.7__py3-none-any.whl → 1.0.8__py3-none-any.whl

meridian/analysis/summarizer.py

@@ -270,6 +270,22 @@ class Summarizer:
   ) -> str:
     """Creates the HTML snippet for the Outcome Contrib card."""
     outcome = self._kpi_or_revenue()
+
+    channel_contrib_area_chart = formatter.ChartSpec(
+        id=summary_text.CHANNEL_CONTRIB_BY_TIME_CHART_ID,
+        description=summary_text.CHANNEL_CONTRIB_BY_TIME_CHART_DESCRIPTION.format(
+            outcome=outcome
+        ),
+        chart_json=media_summary.plot_channel_contribution_area_chart().to_json(),
+    )
+
+    channel_contrib_bump_chart = formatter.ChartSpec(
+        id=summary_text.CHANNEL_CONTRIB_RANK_CHART_ID,
+        description=summary_text.CHANNEL_CONTRIB_RANK_CHART_DESCRIPTION.format(
+            outcome=outcome
+        ),
+        chart_json=media_summary.plot_channel_contribution_bump_chart().to_json(),
+    )
     channel_drivers_chart = formatter.ChartSpec(
         id=summary_text.CHANNEL_DRIVERS_CHART_ID,
         description=summary_text.CHANNEL_DRIVERS_CHART_DESCRIPTION.format(
@@ -308,6 +324,8 @@ class Summarizer:
             channel_drivers_chart,
             spend_outcome_chart,
             outcome_contribution_chart,
+            channel_contrib_area_chart,
+            channel_contrib_bump_chart,
         ],
     )
 
@@ -318,7 +336,7 @@ class Summarizer:
       ascending: bool = False,
   ) -> pd.DataFrame:
     return (
-        media_summary.paid_summary_metrics[metrics]
+        media_summary.get_paid_summary_metrics()[metrics]
         .sel(distribution=c.POSTERIOR, metric=c.MEAN)
         .drop_sel(channel=c.ALL_CHANNELS)
         .to_dataframe()
@@ -334,7 +352,7 @@ class Summarizer:
       ascending: bool = False,
   ) -> pd.DataFrame:
     return (
-        media_summary.paid_summary_metrics[metrics]
+        media_summary.get_paid_summary_metrics()[metrics]
         .sel(distribution=c.POSTERIOR, metric=c.MEDIAN)
         .drop_sel(channel=c.ALL_CHANNELS)
         .to_dataframe()
@@ -479,7 +497,7 @@ class Summarizer:
     rf_channels = reach_frequency.optimal_frequency_data.rf_channel
     assert rf_channels.size > 0
     # This will raise KeyError if not all `rf_channels` can be found in here:
-    rf_channel_spends = media_summary.paid_summary_metrics[c.SPEND].sel(
+    rf_channel_spends = media_summary.get_paid_summary_metrics()[c.SPEND].sel(
         channel=rf_channels
     )
     most_spend_rf_channel = rf_channel_spends.idxmax()

meridian/analysis/summary_text.py

@@ -20,7 +20,9 @@ MODEL_RESULTS_TITLE = 'Marketing Mix Modeling Report'
 MODEL_FIT_CARD_ID = 'model-fit'
 MODEL_FIT_CARD_TITLE = 'Model fit'
 MODEL_FIT_INSIGHTS_FORMAT = """Model fit is a measure of how well your MMM fits
-your current data used to train the model."""
+the data used to train the model. The best model for causal inference may differ
+from the best fitting model, because causal inference models must also estimate
+the unobserved baseline."""
 
 EXPECTED_ACTUAL_OUTCOME_CHART_ID = 'expected-actual-outcome-chart'
 EXPECTED_ACTUAL_OUTCOME_CHART_TITLE = 'Expected {outcome} vs. actual {outcome}'
@@ -42,6 +44,23 @@ CHANNEL_CONTRIB_INSIGHTS_FORMAT = """Your channel contributions help you
 understand what drove your {outcome}. {lead_channels} drove the most overall
 {outcome}."""
 
+CHANNEL_CONTRIB_BY_TIME_CHART_ID = 'channel-contrib-by-time-chart'
+CHANNEL_CONTRIB_BY_TIME_CHART_TITLE = (
+    'Contribution over time by baseline and marketing channels'
+)
+CHANNEL_CONTRIB_BY_TIME_CHART_DESCRIPTION = """Note: This chart shows the
+estimated incremental {outcome} attributed to each channel and the baseline over
+the selected time period. It helps visualize how contributions have changed."""
+
+CHANNEL_CONTRIB_RANK_CHART_ID = 'channel-contrib-rank-chart'
+CHANNEL_CONTRIB_RANK_CHART_TITLE = (
+    'Contribution rank over time by baseline and marketing channels'
+)
+CHANNEL_CONTRIB_RANK_CHART_DESCRIPTION = """Note: This chart shows the relative
+rank of each channel's contribution, including the baseline, based on
+incremental {outcome} at the end of each quarter. Rank 1 represents the highest
+contribution."""
+
 CHANNEL_DRIVERS_CHART_ID = 'channel-drivers-chart'
 CHANNEL_DRIVERS_CHART_TITLE = 'Contribution by baseline and marketing channels'
 CHANNEL_DRIVERS_CHART_DESCRIPTION = """Note: This graphic encompasses all of

meridian/analysis/templates/chart.html.jinja

@@ -28,6 +28,7 @@ limitations under the License.
     const opt = {
       mode: 'vega-lite',
       width: 'container',
+      autosize: { type: 'fit', contains: 'padding' }
     };
     const spec = JSON.parse({{ chart_json|tojson }});
     const chartDiv = document.getElementById('{{ id }}');

meridian/analysis/test_utils.py

@@ -1813,32 +1813,6 @@ MROI_MEDIA_AND_RF_USE_POSTERIOR_BY_REACH = np.array([
         [0.2282, 0.3271, 0.1354, 1.8194, 0.5598],
     ],
 ])
-MROI_MEDIA_AND_RF_NEW_TIMES_DATA = np.array([
-    [
-        [1.4194793, 1.4562954, 0.42110616, 0.46693847, 2.224705],
-        [1.4079778, 1.4408227, 0.42190424, 0.46729392, 2.2684476],
-        [1.4133729, 1.455379, 0.42965493, 0.46928722, 2.2630167],
-        [1.3786758, 1.4669218, 0.4399548, 0.466466, 2.2499163],
-        [1.3667594, 1.4595027, 0.430115, 0.46791035, 2.2718313],
-        [1.3636469, 1.4838042, 0.43774834, 0.46620503, 2.228661],
-        [1.3479362, 1.5108197, 0.42130333, 0.46699694, 2.2348554],
-        [1.3684787, 1.5252702, 0.4246083, 0.4682973, 2.2124028],
-        [1.3500556, 1.5139565, 0.42371163, 0.46387878, 2.2052019],
-        [1.3509449, 1.5152782, 0.42376328, 0.46421173, 2.2056532],
-    ],
-    [
-        [0.19265468, 0.3131754, 0.11835674, 1.7264867, 0.45867893],
-        [0.19271582, 0.31310934, 0.11833327, 1.7265227, 0.45878735],
-        [0.1927399, 0.3133164, 0.11834618, 1.7260367, 0.45711297],
-        [0.19282141, 0.31354108, 0.11858677, 1.724129, 0.45773327],
-        [0.19184875, 0.31352347, 0.11844476, 1.724147, 0.4579198],
-        [0.19213778, 0.3136204, 0.11846119, 1.7245249, 0.4575424],
-        [0.19298446, 0.3144178, 0.11840369, 1.7243268, 0.45999327],
-        [0.19473709, 0.3146425, 0.1182382, 1.7254068, 0.46479526],
-        [0.19482231, 0.31446627, 0.11806685, 1.7242908, 0.46568453],
-        [0.19479823, 0.31506982, 0.11783329, 1.7254248, 0.46606192],
-    ],
-])
 MROI_MEDIA_ONLY_USE_PRIOR = np.array([[
     [1.0740, 1.3019, 0.7984],
     [0.8990, 0.4201, 0.7120],
@@ -2025,32 +1999,6 @@ SAMPLE_ROI_NEW_DATA = np.array([
         [4.11828271, 1.58598067],
     ],
 ])
-ROI_NEW_TIMES_DATA = np.array([
-    [
-        [1.6587073, 1.6422542, 0.5339541, 0.46695647, 2.2247107],
-        [1.6448834, 1.6238545, 0.5349834, 0.46725115, 2.268456],
-        [1.6512104, 1.6399834, 0.54364294, 0.46939594, 2.2630484],
-        [1.6124083, 1.6505513, 0.55594736, 0.46645072, 2.2499137],
-        [1.597995, 1.6415955, 0.5437353, 0.46789172, 2.2718024],
-        [1.5946435, 1.6684949, 0.5538136, 0.46621037, 2.2286298],
-        [1.5757265, 1.6986614, 0.5317232, 0.4669749, 2.2348094],
-        [1.5993005, 1.7154359, 0.535261, 0.4682574, 2.2124054],
-        [1.5770503, 1.7027655, 0.5336275, 0.4639002, 2.2051802],
-        [1.5781163, 1.704347, 0.53381217, 0.4641966, 2.2056897],
-    ],
-    [
-        [0.22893494, 0.54465544, 0.17262405, 1.726445, 0.4586947],
-        [0.22897178, 0.54465, 0.17261527, 1.7264631, 0.45878458],
-        [0.22901268, 0.5448365, 0.17263621, 1.7260604, 0.45709604],
-        [0.22900137, 0.54496145, 0.17280822, 1.7242719, 0.4577607],
-        [0.22795935, 0.54490256, 0.17272605, 1.7241275, 0.4579781],
-        [0.22818668, 0.5450325, 0.1726518, 1.7245975, 0.45759043],
-        [0.22910665, 0.5459215, 0.17257686, 1.7243998, 0.4600002],
-        [0.23114449, 0.54561883, 0.17233835, 1.7253716, 0.46480042],
-        [0.23121877, 0.5451343, 0.17214388, 1.7242151, 0.46571606],
-        [0.23116706, 0.54571074, 0.17181544, 1.7254194, 0.4660912],
-    ],
-])
 SAMPLE_ROI_KPI = np.array([
     [
         [0.4906, 0.3487],
@@ -2617,9 +2565,12 @@ SAMPLE_BASELINE_PCT_OF_CONTRIBUTION_NON_PAID = np.array([
     [-1.204522e-02, -4.079909e02],
     [1.350970e-02, -2.750774e02],
 ])
-ADSTOCK_DECAY_CI_HI = np.array([1.0, 1.0, 0.8295, 0.9728, 0.6880])
-ADSTOCK_DECAY_CI_LO = np.array([1.0, 1.0, 0.8128, 0.6194, 0.6607])
-ADSTOCK_DECAY_MEAN = np.array([1.0, 1.0, 0.8214, 0.8359, 0.6748])
+ADSTOCK_DECAY_CI_HI = np.array([1.0, 1.0, 0.8658, 0.9709, 0.7496])
+ADSTOCK_DECAY_CI_LO = np.array([1.0, 1.0, 0.8328, 0.5749, 0.6936])
+ADSTOCK_DECAY_MEAN = np.array([1.0, 1.0, 0.8493, 0.8630, 0.7215])
+ORGANIC_ADSTOCK_DECAY_CI_HI = np.array([1.0, 0.9636, 0.9291, 0.8962, 0.8650])
+ORGANIC_ADSTOCK_DECAY_CI_LO = np.array([1.0, 0.6623, 0.4394, 0.2920, 0.1944])
+ORGANIC_ADSTOCK_DECAY_MEAN = np.array([1.0, 0.8076, 0.6633, 0.5537, 0.4693])
 HILL_CURVES_CI_HI = np.array([0.0, 0.0, 0.00098, 0.00895, 0.00195])
 HILL_CURVES_CI_LO = np.array([0.0, 0.0, 0.00085, 0.00322, 0.00169])
 HILL_CURVES_MEAN = np.array([0.0, 0.0, 0.00091, 0.00606, 0.00183])
@@ -2864,16 +2815,13 @@ def generate_predictive_accuracy_table(
   shape = [len(metric), len(geo_granularity)]
   dims = [c.METRIC, c.GEO_GRANULARITY]
   coords = {
-      c.METRIC: ([c.METRIC], metric),
-      c.GEO_GRANULARITY: ([c.GEO_GRANULARITY], geo_granularity),
+      c.METRIC: metric,
+      c.GEO_GRANULARITY: geo_granularity,
   }
   if with_holdout:
     shape.append(len(evaluation_set))
     dims.append(c.EVALUATION_SET_VAR)
-    coords[c.EVALUATION_SET_VAR] = (
-        [c.EVALUATION_SET_VAR],
-        evaluation_set,
-    )
+    coords[c.EVALUATION_SET_VAR] = evaluation_set
   np.random.seed(0)
   value = np.random.lognormal(0, 1, size=shape)
   ds = xr.Dataset(
@@ -2969,7 +2917,7 @@ def generate_paid_summary_metrics() -> xr.Dataset:
   )
 
 
-def generate_all_summary_metrics() -> xr.Dataset:
+def generate_all_summary_metrics(aggregate_times: bool = True) -> xr.Dataset:
   """Helper method to generate simulated summary metrics data."""
   channel = (
       [f"ch_{i}" for i in range(3)]
@@ -2981,33 +2929,38 @@ def generate_all_summary_metrics() -> xr.Dataset:
   channel.append(c.ALL_CHANNELS)
   metric = [c.MEAN, c.MEDIAN, c.CI_LO, c.CI_HI]
   distribution = [c.PRIOR, c.POSTERIOR]
+  time = pd.date_range("2023-01-01", freq="W-SUN", periods=5).format(
+      formatter=lambda x: x.strftime("%Y-%m-%d")
+  )
 
   np.random.seed(0)
-  shape = (len(channel), len(metric), len(distribution))
+
+  if aggregate_times:
+    shape = (len(channel), len(metric), len(distribution))
+    dims = [c.CHANNEL, c.METRIC, c.DISTRIBUTION]
+  else:
+    shape = (len(time), len(channel), len(metric), len(distribution))
+    dims = [c.TIME, c.CHANNEL, c.METRIC, c.DISTRIBUTION]
+
   incremental_outcome = np.random.lognormal(10, 1, size=shape)
   effectiveness = np.random.lognormal(1, 1, size=shape)
   pct_of_contribution = np.random.randint(low=0, high=50, size=shape)
 
+  coords = {
+      c.CHANNEL: channel,
+      c.METRIC: metric,
+      c.DISTRIBUTION: distribution,
+  }
+  if not aggregate_times:
+    coords[c.TIME] = time
+
   return xr.Dataset(
       data_vars={
-          c.INCREMENTAL_OUTCOME: (
-              [c.CHANNEL, c.METRIC, c.DISTRIBUTION],
-              incremental_outcome,
-          ),
-          c.PCT_OF_CONTRIBUTION: (
-              [c.CHANNEL, c.METRIC, c.DISTRIBUTION],
-              pct_of_contribution,
-          ),
-          c.EFFECTIVENESS: (
-              [c.CHANNEL, c.METRIC, c.DISTRIBUTION],
-              effectiveness,
-          ),
-      },
-      coords={
-          c.CHANNEL: channel,
-          c.METRIC: metric,
-          c.DISTRIBUTION: distribution,
+          c.INCREMENTAL_OUTCOME: (dims, incremental_outcome),
+          c.PCT_OF_CONTRIBUTION: (dims, pct_of_contribution),
+          c.EFFECTIVENESS: (dims, effectiveness),
       },
+      coords=coords,
       attrs={c.CONFIDENCE_LEVEL: c.DEFAULT_CONFIDENCE_LEVEL},
   )
 
@@ -3063,14 +3016,8 @@ def generate_predictive_accuracy_data(holdout_id: bool = False) -> xr.Dataset:
 
   xr_dims = [c.METRIC, c.GEO_GRANULARITY]
   xr_coords = {
-      c.METRIC: (
-          [c.METRIC],
-          [c.R_SQUARED, c.MAPE, c.WMAPE],
-      ),
-      c.GEO_GRANULARITY: (
-          [c.GEO_GRANULARITY],
-          [c.GEO, c.NATIONAL],
-      ),
+      c.METRIC: [c.R_SQUARED, c.MAPE, c.WMAPE],
+      c.GEO_GRANULARITY: [c.GEO, c.NATIONAL],
   }
   rsquared_arr = [np.random.uniform(0.0, 1.0) for _ in range(2)]
   mape_arr = [np.random.uniform(0.0, 1.0) for _ in range(2)]
@@ -3095,10 +3042,7 @@ def generate_predictive_accuracy_data(holdout_id: bool = False) -> xr.Dataset:
     )
 
     xr_dims.append(c.EVALUATION_SET_VAR)
-    xr_coords[c.EVALUATION_SET_VAR] = (
-        [c.EVALUATION_SET_VAR],
-        list(c.EVALUATION_SET),
-    )
+    xr_coords[c.EVALUATION_SET_VAR] = list(c.EVALUATION_SET)
     xr_data = {c.VALUE: (xr_dims, stacked_total)}
 
   return xr.Dataset(data_vars=xr_data, coords=xr_coords)
@@ -3194,18 +3138,22 @@ def generate_optimal_frequency_data(
 
 def generate_hill_curves_dataframe() -> pd.DataFrame:
   """Helper method to generate simulated hill curve data."""
-  channel_names = [f"ch_{i}" for i in range(3)] + [
-      f"rf_ch_{i}" for i in range(2)
-  ]
+  channel_names = (
+      [f"ch_{i}" for i in range(3)]
+      + [f"rf_ch_{i}" for i in range(2)]
+      + [f"organic_ch_{i}" for i in range(2)]
+  )
   channel_array = []
   channel_type_array = []
-  for i, channel in enumerate(channel_names):
+  for channel_name in channel_names:
     for _ in range(100):
-      channel_array.append(channel)
-      if i <= 3:
+      channel_array.append(channel_name)
+      if channel_name.startswith("ch_"):
         channel_type_array.append(c.MEDIA)
-      else:
+      elif channel_name.startswith("rf_ch_"):
         channel_type_array.append(c.RF)
+      elif channel_name.startswith("organic_ch_"):
+        channel_type_array.append(c.ORGANIC_MEDIA)
 
   np.random.seed(0)
   media_units_array = [