add a streamlit graphical analysis app

examples/app.py

@@ -0,0 +1,166 @@
+# Copyright (c) Microsoft Corporation.
+# Licensed under the MIT License.
+
+"""
+Instruction:
+1. Install streamlit by conda or pip.
+2. qrun the ymal file in the examples folder at least once in order to make a mlruns directory inside examples.
+3. run "streamlit run app.py" and enjoy.
+"""
+
+import itertools
+
+import pandas as pd
+import streamlit as st
+
+import qlib
+from qlib.contrib.report import analysis_model, analysis_position
+from qlib.utils.exceptions import LoadObjectError
+from qlib.workflow import R
+
+
+def _max_width_():
+    # set the app width with a big number in order to adjust wide screens
+    max_width_str = f"max-width: 4000px;"
+    st.markdown(
+        f"""
+    <style>
+    .reportview-container .main .block-container{{
+        {max_width_str}
+    }}
+    </style>    
+    """,
+        unsafe_allow_html=True,
+    )
+
+
+@st.cache
+def _init():
+    # cache the initiation in order to only initiate once
+    qlib.init()
+
+
+_init()
+_max_width_()
+
+experiments = R.list_experiments()
+exp_names = list(experiments.keys())
+selected_exp_name = st.sidebar.selectbox("Please select the experiment", exp_names)
+
+recorders = R.list_recorders(experiment_name=selected_exp_name)
+recorder_names = list(recorders.keys())
+selected_recoder_name = st.sidebar.selectbox(
+    "Please select the recoder", recorder_names
+)
+
+selected_recoder = recorders[selected_recoder_name]
+
+
+@st.cache(max_entries=5, allow_output_mutation=True)
+def get_recorder_artifacts(recoder):
+
+    artifacts = dict()
+    artifacts["params"] = recoder.list_params()
+    artifacts["metrics"] = recoder.list_metrics()
+    artifacts["tags"] = recoder.list_tags()
+    try:
+        artifacts["report_normal_df"] = recoder.load_object(
+            "portfolio_analysis/report_normal.pkl"
+        )
+    except LoadObjectError:
+        pass
+    try:
+        artifacts["analysis_df"] = recoder.load_object(
+            "portfolio_analysis/port_analysis.pkl"
+        )
+    except LoadObjectError:
+        pass
+    try:
+        pred_df = recoder.load_object("pred.pkl")
+        label_df = recoder.load_object("label.pkl")
+        label_df.columns = ["label"]
+        artifacts["pred_label"] = pd.concat(
+            [label_df, pred_df], axis=1, sort=True
+        ).reindex(label_df.index)
+    except LoadObjectError:
+        pass
+    try:
+        artifacts["positions"] = selected_recoder.load_object(
+            "portfolio_analysis/positions_normal.pkl"
+        )
+    except LoadObjectError:
+        pass
+
+    return artifacts
+
+
+loaded_artifacts = get_recorder_artifacts(selected_recoder)
+
+params = loaded_artifacts.get("params")
+metrics = loaded_artifacts.get("metrics")
+tags = loaded_artifacts.get("tags")
+
+st.write(params)
+st.write(metrics)
+st.write(tags)
+
+positions = loaded_artifacts.get("positions")
+
+if positions is not None:
+    dates = list(positions.keys())
+    col1, col2 = st.beta_columns(2)
+    with col1:
+        try:
+            date = st.date_input("Please select a date", value=dates[0])
+            date = pd.Timestamp(date)
+            position_at_date = pd.DataFrame(positions[date])
+
+            st.dataframe(
+                position_at_date.loc[
+                    :, ~position_at_date.columns.isin(["cash", "today_account_value"])
+                ].T.style.background_gradient("Reds")
+            )
+        except (ValueError, KeyError):
+            st.write("This is not a valid date")
+
+    generator = (value.keys() for key, value in positions.items())
+    stocks = list(set(itertools.chain.from_iterable(generator)))
+    stocks.sort()
+
+    with col2:
+        selected_stock = st.selectbox("Please select a stock", stocks)
+        try:
+            weight = pd.Series(
+                (
+                    positions[d].get(selected_stock, {"weight": 0.0})["weight"]
+                    for d in dates
+                ),
+                index=pd.DatetimeIndex(dates),
+                name="weight",
+                dtype="float",
+            )
+
+            st.line_chart(weight)
+        except TypeError:
+            st.write("This is not a valid stock code")
+
+
+report_normal_df = loaded_artifacts.get("report_normal_df")
+analysis_df = loaded_artifacts.get("analysis_df")
+
+if report_normal_df is not None:
+    for fig in analysis_position.report_graph(report_normal_df, show_notebook=False):
+        st.plotly_chart(fig, use_container_width=True)
+    if analysis_df is not None:
+        for fig in analysis_position.risk_analysis_graph(
+            analysis_df, report_normal_df, show_notebook=False
+        ):
+            st.plotly_chart(fig, use_container_width=True)
+
+pred_label = loaded_artifacts.get("pred_label")
+
+if pred_label is not None:
+    for fig in analysis_position.score_ic_graph(pred_label, show_notebook=False):
+        st.plotly_chart(fig, use_container_width=True)
+    for fig in analysis_model.model_performance_graph(pred_label, show_notebook=False):
+        st.plotly_chart(fig, use_container_width=True)