Create streamlit_app.py

wise_pizza/streamlit_app.py

@@ -0,0 +1,342 @@
+"""An example of Streamlit leveraging Wise pizza."""
+# Streamlit part was initially developed by https://github.com/agusfigueroa-htg
+
+import streamlit as st
+import os, sys
+import pandas as pd
+from wise_pizza.explain import (
+    explain_levels,
+    explain_changes_in_totals,
+    explain_changes_in_average,
+)
+
+def run_streamlit_app():
+
+    # False if you want nice interactive plots
+    # True if you want static plots (Doesn't work on all platforms yet)
+    plot_is_static = False
+
+
+    # SETTING PAGE CONFIG TO WIDE MODE AND ADDING A TITLE AND FAVICON
+    st.set_page_config(layout="wide", page_title="Wise Pizza", page_icon=":pizza:")
+
+    st.title("Wise Pizza powered by Streamlit")
+    st.text(
+        "Only categorical columns are accepted, bucket the numeric ones if you wanna use those"
+    )
+
+    # upload the file from the computer
+    def load_data_upload():
+        uploaded_file = st.file_uploader("Choose a file")
+        if not uploaded_file:
+            st.warning("Please input a dataset.")
+            data = pd.DataFrame()
+            st.stop()
+        else:
+            st.success("Dataset inputted.")
+            data = pd.read_csv(uploaded_file)
+        return data
+
+
+    on = st.toggle("Use sample data from Github")
+    url_data = r"https://raw.githubusercontent.com/transferwise/wise-pizza/main/data/synth_data.csv"
+
+    # select the datasource, either local or from github
+    if on:
+        st.write(f"Downloading data from {url_data}!")
+        df = pd.read_csv(url_data)
+    else:
+        df = load_data_upload()
+
+    # show dataset preview
+    st.text("Table preview")
+    st.table(df.head(10))
+
+    # ask the user for relevant dimensions
+    dims = st.multiselect(
+        "Select the dimensions you want to include in the analysis",
+        df.select_dtypes(exclude=["number"]).columns.tolist(),
+    )
+
+    # ask the user via streamlit to select if they want to run a comparison between subgroups or not
+    flag_comparison = st.toggle(
+        "I want to run a comparison between two subgroups in my data"
+    )
+
+    # return columns that are candidate for comparison
+    def flag_columns(df):
+        binary_columns = df.columns[df.nunique() == 2].tolist()
+        if not binary_columns:
+            st.warning(
+                "No column in the dataset is binary, no comparison can be carried out"
+            )
+            st.stop()
+        return binary_columns
+
+
+    if flag_comparison:
+        # calculate binary columns
+        binary_columns = flag_columns(df)
+
+        # ask the user via streamlit for the flag column
+        flag_column = st.selectbox(
+            "What is the flag column of your dataset that defines the two subgroups?",
+            binary_columns,
+            index=None,
+            placeholder="Select flag column column",
+        )
+        st.write("You selected:", flag_column)
+
+        # wait until flag column is added
+        if not flag_column:
+            st.stop()
+
+        # calculate unique flags in the dataset
+        flags = sorted(df[flag_column].unique())
+
+        # allow users to define what's "old" and "new" in the comparison
+        flags_option = st.selectbox(
+            "Which value in your flag column belongs to group A?", (flags)
+        )
+
+
+    # ask the user via streamlit for the target column
+    totals = st.selectbox(
+        "Name of column that contains totals per segment (e.g. GMV/revenue)",
+        # display only numerical columns
+        df.select_dtypes(include=["number"]).columns.tolist(),
+        index=None,
+        placeholder="Select target column",
+    )
+    st.write("You selected:", totals)
+
+    # ask the user via streamlit for the size column
+    size = st.selectbox(
+        "Name of column containing segment size (e.g. number of users/number of transactions)",
+        # display only numerical columns
+        df.select_dtypes(include=["number"]).columns.tolist(),
+        index=None,
+        placeholder="Select volume column",
+    )
+    st.write("You selected:", size)
+
+    if not totals or not size or not dims:
+        st.warning("Please input all fields above.")
+        st.stop()
+
+    if not flag_comparison:
+        st.subheader("Finding the juiciest slices", divider="rainbow")
+        st.text("Find segments whose average is most different from the global one")
+        ## finding juiciest slices
+        solver = st.selectbox("Select a solver:", ("lasso", "lp"))
+        min_segments = st.number_input(
+            "Min segments: Minimum number of segments to find",
+            min_value=1,
+            max_value=200,
+            value=20,
+            step=1,
+            key="min_segments",
+        )
+        max_segments = st.number_input(
+            "Max segments: Maximum number of segments to find, defaults to min_segments",
+            min_value=1,
+            max_value=200,
+            value=20,
+            step=1,
+            key="max_segments",
+        )
+        min_depth = st.number_input(
+            "Min depth: Minimum number of dimension to constrain in segment definition",
+            min_value=1,
+            max_value=10,
+            value=1,
+            step=1,
+            key="min_depth",
+        )
+        max_depth = st.number_input(
+            "Max depth: Maximum number of dimension to constrain in segment definition",
+            min_value=1,
+            max_value=10,
+            value=2,
+            step=1,
+            key="max_depth",
+        )
+        cluster_values = st.selectbox(
+            "Cluster values: In addition to single-value slices, consider slices that consist of a group of segments from the same dimension with similar naive averages",
+            (True, False),
+        )
+
+        width = st.number_input(
+            "Width of the plot",
+            min_value=100,
+            max_value=10000,
+            value=500,
+            step=1,
+            key="width",
+        )
+        height = st.number_input(
+            "Height of the plot",
+            min_value=100,
+            max_value=10000,
+            value=500,
+            step=1,
+            key="height",
+        )
+        apply_button = st.button("Apply")
+        if apply_button:
+            sf = explain_levels(
+                df=df,
+                dims=dims,
+                total_name=totals,
+                size_name=size,
+                min_depth=min_depth,
+                max_depth=max_depth,
+                min_segments=min_segments,
+                max_segments=max_segments,
+                solver=solver,
+                cluster_values=cluster_values,
+            )
+            plot_sf = sf.plot(width=width, height=height, return_fig=True)
+            if not cluster_values:
+                plot_sf_0 = plot_sf
+                st.plotly_chart(plot_sf_0, use_container_width=True)
+            else:
+                plot_sf_0 = plot_sf[0]
+                st.plotly_chart(plot_sf_0, use_container_width=True)
+                plot_sf_1 = plot_sf[1]
+                st.plotly_chart(plot_sf_1, use_container_width=True)
+
+    if flag_comparison:
+        st.subheader("Analysing differences", divider="rainbow")
+        st.text(
+            "This section does compare the two groups defined by the flag. Old total is the group A you selected in the dropdown"
+        )
+        st.text(
+            "Explain changes in totals: Let's look for segments that experience the largest change in the totals from previous period."
+        )
+        st.text(
+            "Explain changes in average: Sometimes, rather than explaining the change in totals from one period to the next, one wishes to explain a change in averages"
+        )
+
+        data = df[df[flag_column] != flags_option]  # take the group to compare to
+        pre_data = df[df[flag_column] == flags_option]  # take the group to be compared
+
+        comparison_dimensions = list(filter(lambda x: x != flag_column, dims))
+
+        explain = st.selectbox(
+            "Select a function:",
+            ("explain changes in totals", "explain changes in average"),
+        )
+        solver = st.selectbox("Select a solver:", ("lasso", "lp"))
+        min_segments = st.number_input(
+            "Min segments: Minimum number of segments to find",
+            min_value=1,
+            max_value=200,
+            value=20,
+            step=1,
+            key="min_segments",
+        )
+        max_segments = st.number_input(
+            "Max segments: Maximum number of segments to find, defaults to min_segments",
+            min_value=1,
+            max_value=200,
+            value=20,
+            step=1,
+            key="max_segments",
+        )
+        min_depth = st.number_input(
+            "Min depth: Minimum number of dimension to constrain in segment definition",
+            min_value=1,
+            max_value=10,
+            value=1,
+            step=1,
+            key="min_depth",
+        )
+        max_depth = st.number_input(
+            "Max depth: Maximum number of dimension to constrain in segment definition",
+            min_value=1,
+            max_value=10,
+            value=2,
+            step=1,
+            key="max_depth",
+        )
+        cluster_values = st.selectbox(
+            "Cluster values: In addition to single-value slices, consider slices that consist of a group of segments from the same dimension with similar naive averages",
+            (True, False),
+        )
+        how = st.selectbox(
+            "Select a method:", ("totals", "split_fits", "extra_dim", "force_dim")
+        )
+        st.text(
+            '"totals": to only decompose segment totals (ignoring size vs average contribution)'
+        )
+        st.text(
+            '"split_fits": to separately decompose contribution of size changes and average changes'
+        )
+        st.text(
+            '"extra_dim": to treat size vs average change contribution as an additional dimension'
+        )
+        st.text(
+            '"force_dim": like extra_dim, but each segment must contain a Change_from constraint'
+        )
+        width = st.number_input(
+            "Width of the plot",
+            min_value=100,
+            max_value=10000,
+            value=500,
+            step=1,
+            key="width",
+        )
+        height = st.number_input(
+            "Height of the plot",
+            min_value=100,
+            max_value=10000,
+            value=500,
+            step=1,
+            key="height",
+        )
+        changes_apply_button = st.button("Get results")
+        ## running explain calculations
+        if changes_apply_button:
+            if explain == "explain changes in totals":
+                sf1 = explain_changes_in_totals(
+                    df1=pre_data,
+                    df2=data,
+                    dims=comparison_dimensions,
+                    total_name=totals,
+                    size_name=size,
+                    min_depth=min_depth,
+                    max_depth=max_depth,
+                    min_segments=min_segments,
+                    max_segments=max_segments,
+                    solver=solver,
+                    cluster_values=cluster_values,
+                    how=how,
+                )
+            else:
+                sf1 = explain_changes_in_average(
+                    df1=pre_data,
+                    df2=data,
+                    dims=comparison_dimensions,
+                    total_name=totals,
+                    size_name=size,
+                    min_depth=min_depth,
+                    max_depth=max_depth,
+                    min_segments=min_segments,
+                    max_segments=max_segments,
+                    solver=solver,
+                    cluster_values=cluster_values,
+                    how=how,
+                )
+            plot_sf1 = sf1.plot(width=width, height=height, return_fig=True)
+            if not cluster_values:
+                plot_sf_0 = plot_sf1
+                st.plotly_chart(plot_sf_0, use_container_width=True)
+            else:
+                plot_sf_0 = plot_sf1[0]
+                st.plotly_chart(plot_sf_0, use_container_width=True)
+                plot_sf_1 = plot_sf1[1]
+                st.plotly_chart(plot_sf_1, use_container_width=True)
+
+if __name__ == "__main__":
+    run_streamlit_app()