Merge pull request #61 from hummingbot/feat/backtesting-frontend-v2

Feat/backtesting frontend v2

pages/backtest_manager/analyze.py

@@ -1,4 +1,156 @@
+import constants
+import os
+import json
+import streamlit as st
+
+from quants_lab.strategy.strategy_analysis import StrategyAnalysis
+from utils.graphs import BacktestingGraphs
+from utils.optuna_database_manager import OptunaDBManager
+from utils.os_utils import load_directional_strategies
 from utils.st_utils import initialize_st_page
 
-
 initialize_st_page(title="Analyze", icon="🔬", initial_sidebar_state="collapsed")
+
+
+@st.cache_resource
+def get_databases():
+    sqlite_files = [db_name for db_name in os.listdir("data/backtesting") if db_name.endswith(".db")]
+    databases_list = [OptunaDBManager(db) for db in sqlite_files]
+    databases_dict = {database.db_name: database for database in databases_list}
+    return [x.db_name for x in databases_dict.values() if x.status == 'OK']
+
+
+def initialize_session_state_vars():
+    if "strategy_params" not in st.session_state:
+        st.session_state.strategy_params = {}
+    if "backtesting_params" not in st.session_state:
+        st.session_state.backtesting_params = {}
+
+
+initialize_session_state_vars()
+dbs = get_databases()
+if not dbs:
+    st.warning("We couldn't find any Optuna database.")
+    selected_db_name = None
+    selected_db = None
+else:
+    # Select database from selectbox
+    selected_db = st.selectbox("Select your database:", dbs)
+    # Instantiate database manager
+    opt_db = OptunaDBManager(selected_db)
+    # Load studies
+    studies = opt_db.load_studies()
+    # Choose study
+    study_selected = st.selectbox("Select a study:", studies.keys())
+    # Filter trials from selected study
+    merged_df = opt_db.merged_df[opt_db.merged_df["study_name"] == study_selected]
+    bt_graphs = BacktestingGraphs(merged_df)
+    # Show and compare all of the study trials
+    st.plotly_chart(bt_graphs.pnl_vs_maxdrawdown(), use_container_width=True)
+    # Get study trials
+    trials = studies[study_selected]
+    # Choose trial
+    trial_selected = st.selectbox("Select a trial to backtest", list(trials.keys()))
+    trial = trials[trial_selected]
+    # Transform trial config in a dictionary
+    trial_config = json.loads(trial["config"])
+
+    # Strategy parameters section
+    st.write("## Strategy parameters")
+    # Load strategies (class, config, module)
+    strategies = load_directional_strategies(constants.DIRECTIONAL_STRATEGIES_PATH)
+    # Select strategy
+    strategy = strategies[trial_config["name"]]
+    # Get field schema
+    field_schema = strategy["config"].schema()["properties"]
+    c1, c2 = st.columns([5, 1])
+    # Render every field according to schema
+    with c1:
+        columns = st.columns(4)
+        column_index = 0
+        for field_name, properties in field_schema.items():
+            field_type = properties["type"]
+            field_value = trial_config[field_name]
+            with columns[column_index]:
+                if field_type in ["number", "integer"]:
+                    field_value = st.number_input(field_name,
+                                                  value=field_value,
+                                                  min_value=properties.get("minimum"),
+                                                  max_value=properties.get("maximum"),
+                                                  key=field_name)
+                elif field_type == "string":
+                    field_value = st.text_input(field_name, value=field_value)
+                elif field_type == "boolean":
+                    # TODO: Add support for boolean fields in optimize tab
+                    field_value = st.checkbox(field_name, value=field_value)
+                else:
+                    raise ValueError(f"Field type {field_type} not supported")
+                try:
+                    st.session_state["strategy_params"][field_name] = field_value
+                except KeyError as e:
+                    pass
+            column_index = (column_index + 1) % 4
+    with c2:
+        add_positions = st.checkbox("Add positions", value=True)
+        add_volume = st.checkbox("Add volume", value=True)
+        add_pnl = st.checkbox("Add PnL", value=True)
+
+    # Backtesting parameters section
+    st.write("## Backtesting parameters")
+    # Get every trial params
+    # TODO: Filter only from selected study
+    backtesting_configs = opt_db.load_params()
+    # Get trial backtesting params
+    backtesting_params = backtesting_configs[trial_selected]
+    col1, col2, col3 = st.columns(3)
+    with col1:
+        selected_order_amount = st.number_input("Order amount",
+                                                value=50.0,
+                                                min_value=0.1,
+                                                max_value=999999999.99)
+        selected_leverage = st.number_input("Leverage",
+                                            value=10,
+                                            min_value=1,
+                                            max_value=200)
+    with col2:
+        selected_initial_portfolio = st.number_input("Initial portfolio",
+                                                     value=10000.00,
+                                                     min_value=1.00,
+                                                     max_value=999999999.99)
+        selected_time_limit = st.number_input("Time Limit",
+                                              value=60 * 60 * backtesting_params["time_limit"]["param_value"],
+                                              min_value=60 * 60 * float(backtesting_params["time_limit"]["low"]),
+                                              max_value=60 * 60 * float(backtesting_params["time_limit"]["high"]))
+    with col3:
+        selected_tp_multiplier = st.number_input("Take Profit Multiplier",
+                                                 value=backtesting_params["take_profit_multiplier"]["param_value"],
+                                                 min_value=backtesting_params["take_profit_multiplier"]["low"],
+                                                 max_value=backtesting_params["take_profit_multiplier"]["high"])
+        selected_sl_multiplier = st.number_input("Stop Loss Multiplier",
+                                                 value=backtesting_params["stop_loss_multiplier"]["param_value"],
+                                                 min_value=backtesting_params["stop_loss_multiplier"]["low"],
+                                                 max_value=backtesting_params["stop_loss_multiplier"]["high"])
+
+    if st.button("Run Backtesting!"):
+        config = strategy["config"](**st.session_state["strategy_params"])
+        strategy = strategy["class"](config=config)
+        try:
+            market_data, positions = strategy.run_backtesting(
+                order_amount=selected_order_amount,
+                leverage=selected_order_amount,
+                initial_portfolio=selected_initial_portfolio,
+                take_profit_multiplier=selected_tp_multiplier,
+                stop_loss_multiplier=selected_sl_multiplier,
+                time_limit=selected_time_limit,
+                std_span=None,
+            )
+            strategy_analysis = StrategyAnalysis(
+                positions=positions,
+                candles_df=market_data,
+            )
+            metrics_container = bt_graphs.get_trial_metrics(strategy_analysis,
+                                                            add_positions=add_positions,
+                                                            add_volume=add_volume,
+                                                            add_pnl=add_pnl)
+        except FileNotFoundError:
+            st.warning(f"The requested candles could not be found.")

quants_lab/strategy/strategy_analysis.py

@@ -193,6 +193,15 @@ class StrategyAnalysis:
         time_diff_minutes = (pd.to_datetime(self.positions['close_time']) - self.positions['timestamp']).dt.total_seconds() / 60
         return time_diff_minutes.mean()
 
+    def start_date(self):
+        return self.candles_df.timestamp.min()
+
+    def end_date(self):
+        return self.candles_df.timestamp.max()
+
+    def avg_profit(self):
+        return self.positions.ret_usd.mean()
+
     def text_report(self):
         return f"""
 Strategy Performance Report:
@@ -207,3 +216,16 @@ Strategy Performance Report:
     - Duration: {self.duration_in_minutes() / 60:,.2f} Hours
     - Average Trade Duration: {self.avg_trading_time_in_minutes():,.2f} minutes
     """
+
+    def pnl_over_time(self):
+        fig = go.Figure()
+        fig.add_trace(go.Scatter(name="PnL Over Time",
+                                 x=self.positions.index,
+                                 y=self.positions.ret_usd.cumsum()))
+        # Update layout with the required attributes
+        fig.update_layout(
+            title="PnL Over Time",
+            xaxis_title="N° Position",
+            yaxis=dict(title="Net PnL USD", side="left", showgrid=False),
+        )
+        return fig

utils/file_templates.py

@@ -11,6 +11,7 @@ from quants_lab.strategy.directional_strategy_base import DirectionalStrategyBas
 
 
 class {strategy_config_cls_name}(BaseModel):
+    name: str = "{strategy_cls_name.lower()}"
     exchange: str = Field(default="binance_perpetual")
     trading_pair: str = Field(default="ETH-USDT")
     interval: str = Field(default="1h")

utils/graphs.py

@@ -4,6 +4,7 @@ import pandas_ta as ta  # noqa: F401
 import streamlit as st
 
 from utils.data_manipulation import StrategyData, SingleMarketStrategyData
+from quants_lab.strategy.strategy_analysis import StrategyAnalysis
 import plotly.graph_objs as go
 
 
@@ -248,3 +249,97 @@ class CandlesGraph:
         merged_df["trade_pnl_continuos"] = merged_df["unrealized_trade_pnl"] + merged_df["cum_net_amount"] * merged_df["close"]
         merged_df["net_pnl_continuos"] = merged_df["trade_pnl_continuos"] - merged_df["cum_fees_in_quote"]
         return merged_df
+
+
+class BacktestingGraphs:
+    def __init__(self, study_df: pd.DataFrame):
+        self.study_df = study_df
+
+    def pnl_vs_maxdrawdown(self):
+        fig = go.Figure()
+        fig.add_trace(go.Scatter(name="Pnl vs Max Drawdown",
+                                 x=-100 * self.study_df["max_drawdown_pct"],
+                                 y=100 * self.study_df["net_profit_pct"],
+                                 mode="markers",
+                                 text=None,
+                                 hovertext=self.study_df["hover_text"]))
+        fig.update_layout(
+            title="PnL vs Max Drawdown",
+            xaxis_title="Max Drawdown [%]",
+            yaxis_title="Net Profit [%]",
+            height=800
+        )
+        fig.data[0].text = []
+        return fig
+
+    @staticmethod
+    def get_trial_metrics(strategy_analysis: StrategyAnalysis,
+                          add_volume: bool = True,
+                          add_positions: bool = True,
+                          add_pnl: bool = True):
+        """Isolated method because it needs to be called from analyze and simulate pages"""
+        metrics_container = st.container()
+        with metrics_container:
+            col1, col2 = st.columns(2)
+            with col1:
+                st.subheader("🏦 Market")
+            with col2:
+                st.subheader("📋 General stats")
+            col1, col2, col3, col4 = st.columns(4)
+            with col1:
+                st.metric("Exchange", st.session_state["strategy_params"]["exchange"])
+            with col2:
+                st.metric("Trading Pair", st.session_state["strategy_params"]["trading_pair"])
+            with col3:
+                st.metric("Start date", strategy_analysis.start_date().strftime("%Y-%m-%d %H:%M"))
+                st.metric("End date", strategy_analysis.end_date().strftime("%Y-%m-%d %H:%M"))
+            with col4:
+                st.metric("Duration (hours)", f"{strategy_analysis.duration_in_minutes() / 60:.2f}")
+                st.metric("Price change", st.session_state["strategy_params"]["trading_pair"])
+            st.subheader("📈 Performance")
+            col1, col2, col3, col4, col5, col6, col7, col8 = st.columns(8)
+            with col1:
+                st.metric("Net PnL USD",
+                          f"{strategy_analysis.net_profit_usd():.2f}",
+                          delta=f"{100 * strategy_analysis.net_profit_pct():.2f}%",
+                          help="The overall profit or loss achieved.")
+            with col2:
+                st.metric("Total positions",
+                          f"{strategy_analysis.total_positions()}",
+                          help="The total number of closed trades, winning and losing.")
+            with col3:
+                st.metric("Accuracy",
+                          f"{100 * (len(strategy_analysis.win_signals()) / strategy_analysis.total_positions()):.2f} %",
+                          help="The percentage of winning trades, the number of winning trades divided by the"
+                               " total number of closed trades")
+            with col4:
+                st.metric("Profit factor",
+                          f"{strategy_analysis.profit_factor():.2f}",
+                          help="The amount of money the strategy made for every unit of money it lost, "
+                               "gross profits divided by gross losses.")
+            with col5:
+                st.metric("Max Drawdown",
+                          f"{strategy_analysis.max_drawdown_usd():.2f}",
+                          delta=f"{100 * strategy_analysis.max_drawdown_pct():.2f}%",
+                          help="The greatest loss drawdown, i.e., the greatest possible loss the strategy had compared "
+                               "to its highest profits")
+            with col6:
+                st.metric("Avg Profit",
+                          f"{strategy_analysis.avg_profit():.2f}",
+                          help="The sum of money gained or lost by the average trade, Net Profit divided by "
+                               "the overall number of closed trades.")
+            with col7:
+                st.metric("Avg Minutes",
+                          f"{strategy_analysis.avg_trading_time_in_minutes():.2f}",
+                          help="The average number of minutes that elapsed during trades for all closed trades.")
+            with col8:
+                st.metric("Sharpe Ratio",
+                          f"{strategy_analysis.sharpe_ratio():.2f}",
+                          help="The Sharpe ratio is a measure that quantifies the risk-adjusted return of an investment"
+                               " or portfolio. It compares the excess return earned above a risk-free rate per unit of"
+                               " risk taken.")
+
+            st.plotly_chart(strategy_analysis.pnl_over_time(), use_container_width=True)
+            strategy_analysis.create_base_figure(volume=add_volume, positions=add_positions, trade_pnl=add_pnl)
+            st.plotly_chart(strategy_analysis.figure(), use_container_width=True)
+        return metrics_container

utils/optuna_database_manager.py

@@ -0,0 +1,297 @@
+import os
+import json
+
+import pandas as pd
+from sqlalchemy import create_engine
+from sqlalchemy.orm import sessionmaker
+
+from utils.data_manipulation import StrategyData
+
+
+class OptunaDBManager:
+    def __init__(self, db_name):
+        self.db_name = db_name
+        self.db_path = f'sqlite:///{os.path.join("data/backtesting", db_name)}'
+        self.engine = create_engine(self.db_path, connect_args={'check_same_thread': False})
+        self.session_maker = sessionmaker(bind=self.engine)
+
+    @property
+    def status(self):
+        try:
+            with self.session_maker() as session:
+                query = 'SELECT * FROM trials WHERE state = "COMPLETE"'
+                completed_trials = pd.read_sql_query(query, session.connection())
+            if len(completed_trials) > 0:
+                # TODO: improve error handling, think what to do with other cases
+                return "OK"
+            else:
+                return "No records found in the trials table with completed state"
+        except Exception as e:
+            return f"Error: {str(e)}"
+
+    @property
+    def tables(self):
+        return self._get_tables()
+
+    def _get_tables(self):
+        try:
+            with self.session_maker() as session:
+                query = "SELECT name FROM sqlite_master WHERE type='table';"
+                tables = pd.read_sql_query(query, session.connection())
+            return tables["name"].tolist()
+        except Exception as e:
+            return f"Error: {str(e)}"
+
+    @property
+    def trials(self):
+        return self._get_trials_table()
+
+    def _get_trials_table(self):
+        try:
+            with self.session_maker() as session:
+                df = pd.read_sql_query("SELECT * FROM trials", session.connection())
+            return df
+        except Exception as e:
+            return f"Error: {str(e)}"
+
+    @property
+    def studies(self):
+        return self._get_studies_table()
+
+    def _get_studies_table(self):
+        try:
+            with self.session_maker() as session:
+                df = pd.read_sql_query("SELECT * FROM studies", session.connection())
+            return df
+        except Exception as e:
+            return f"Error: {str(e)}"
+
+    @property
+    def trial_params(self):
+        return self._get_trial_params_table()
+
+    def _get_trial_params_table(self):
+        try:
+            with self.session_maker() as session:
+                df = pd.read_sql_query("SELECT * FROM trial_params", session.connection())
+            return df
+        except Exception as e:
+            return f"Error: {str(e)}"
+
+    @property
+    def trial_values(self):
+        return self._get_trial_values_table()
+
+    def _get_trial_values_table(self):
+        try:
+            with self.session_maker() as session:
+                df = pd.read_sql_query("SELECT * FROM trial_values", session.connection())
+            return df
+        except Exception as e:
+            return f"Error: {str(e)}"
+
+    @property
+    def trial_system_attributes(self):
+        return self._get_trial_system_attributes_table()
+
+    def _get_trial_system_attributes_table(self):
+        try:
+            with self.session_maker() as session:
+                df = pd.read_sql_query("SELECT * FROM trial_system_attributes", session.connection())
+            return df
+        except Exception as e:
+            return f"Error: {str(e)}"
+
+    @property
+    def trial_system_attributes(self):
+        return self._get_trial_system_attributes_table()
+
+    def _get_trial_system_attributes_table(self):
+        try:
+            with self.session_maker() as session:
+                df = pd.read_sql_query("SELECT * FROM trial_system_attributes", session.connection())
+            return df
+        except Exception as e:
+            return f"Error: {str(e)}"
+
+    @property
+    def version_info(self):
+        return self._get_version_info_table()
+
+    def _get_version_info_table(self):
+        try:
+            with self.session_maker() as session:
+                df = pd.read_sql_query("SELECT * FROM version_info", session.connection())
+            return df
+        except Exception as e:
+            return f"Error: {str(e)}"
+
+    @property
+    def study_directions(self):
+        return self._get_study_directions_table()
+
+    def _get_study_directions_table(self):
+        try:
+            with self.session_maker() as session:
+                df = pd.read_sql_query("SELECT * FROM study_directions", session.connection())
+            return df
+        except Exception as e:
+            return f"Error: {str(e)}"
+
+    @property
+    def study_user_attributes(self):
+        return self._get_study_user_attributes_table()
+
+    def _get_study_user_attributes_table(self):
+        try:
+            with self.session_maker() as session:
+                df = pd.read_sql_query("SELECT * FROM study_user_attributes", session.connection())
+            return df
+        except Exception as e:
+            return f"Error: {str(e)}"
+
+    @property
+    def study_system_attributes(self):
+        return self._get_study_system_attributes_table()
+
+    def _get_study_system_attributes_table(self):
+        try:
+            with self.session_maker() as session:
+                df = pd.read_sql_query("SELECT * FROM study_system_attributes", session.connection())
+            return df
+        except Exception as e:
+            return f"Error: {str(e)}"
+
+    @property
+    def trial_user_attributes(self):
+        return self._get_trial_user_attributes_table()
+
+    def _get_trial_user_attributes_table(self):
+        try:
+            with self.session_maker() as session:
+                df = pd.read_sql_query("SELECT * FROM trial_user_attributes", session.connection())
+            return df
+        except Exception as e:
+            return f"Error: {str(e)}"
+
+    @property
+    def trial_intermediate_values(self):
+        return self._get_trial_intermediate_values_table()
+
+    def _get_trial_intermediate_values_table(self):
+        try:
+            with self.session_maker() as session:
+                df = pd.read_sql_query("SELECT * FROM trial_intermediate_values", session.connection())
+            return df
+        except Exception as e:
+            return f"Error: {str(e)}"
+
+    @property
+    def trial_heartbeats(self):
+        return self._get_trial_heartbeats_table()
+
+    def _get_trial_heartbeats_table(self):
+        try:
+            with self.session_maker() as session:
+                df = pd.read_sql_query("SELECT * FROM trial_heartbeats", session.connection())
+            return df
+        except Exception as e:
+            return f"Error: {str(e)}"
+
+    @property
+    def alembic_version(self):
+        return self._get_alembic_version_table()
+
+    def _get_alembic_version_table(self):
+        try:
+            with self.session_maker() as session:
+                df = pd.read_sql_query("SELECT * FROM alembic_version", session.connection())
+            return df
+        except Exception as e:
+            return f"Error: {str(e)}"
+
+    @property
+    def merged_df(self):
+        return self._get_merged_df()
+
+    @staticmethod
+    def _add_hovertext(x):
+        summary_label = (f"<b>Trial ID: {x['trial_id']}</b><br>"
+                         f"<b>Study: {x['study_name']}</b><br>"
+                         f"--------------------<br>"
+                         f"Accuracy: {100 * x['accuracy']:.2f} %<br>"
+                         f"Avg Trading Time in Hours: {x['avg_trading_time_in_hours']:.2f}<br>"
+                         f"Duration in Hours: {x['duration_in_hours']:.2f}<br>"
+                         f"Loss Signals: {x['loss_signals']}<br>"
+                         f"Max Drawdown [%]: {100 * x['max_drawdown_pct']:.2f} %<br>"
+                         f"Max Drawdown [USD]: $ {x['max_drawdown_usd']:.2f}<br>"
+                         f"Net Profit [%]: {100 * x['net_profit_pct']:.2f} %<br>"
+                         f"Net Profit [$]: $ {x['net_profit_usd']:.2f}<br>"
+                         f"Profit Factor: {x['profit_factor']:.2f}<br>"
+                         f"Sharpe Ratio: {x['sharpe_ratio']:.4f}<br>"
+                         f"Total Positions: {x['total_positions']}<br>"
+                         f"Win Signals: {x['win_signals']}<br>"
+                         f"Trial value: {x['value']}<br>"
+                         f"Direction: {x['direction']}<br>"
+                         )
+        return summary_label
+
+    def _get_merged_df(self):
+        float_cols = ["accuracy", "avg_trading_time_in_hours", "duration_in_hours", "max_drawdown_pct", "max_drawdown_usd",
+                      "net_profit_pct", "net_profit_usd", "profit_factor", "sharpe_ratio", "value"]
+        int_cols = ["loss_signals", "total_positions", "win_signals"]
+        merged_df = self.trials\
+            .merge(self.studies, on="study_id")\
+            .merge(pd.pivot(self.trial_user_attributes, index="trial_id", columns="key", values="value_json"),
+                   on="trial_id")\
+            .merge(self.trial_values, on="trial_id")\
+            .merge(self.study_directions, on="study_id")
+        merged_df[float_cols] = merged_df[float_cols].astype("float")
+        merged_df[int_cols] = merged_df[int_cols].astype("int64")
+        merged_df["hover_text"] = merged_df.apply(self._add_hovertext, axis=1)
+        return merged_df
+
+    def load_studies(self):
+        df = self.merged_df
+        study_name_col = 'study_name'
+        trial_id_col = 'trial_id'
+        nested_dict = {}
+        for _, row in df.iterrows():
+            study_name = row[study_name_col]
+            trial_id = row[trial_id_col]
+            data_dict = row.drop([study_name_col, trial_id_col]).to_dict()
+            if study_name not in nested_dict:
+                nested_dict[study_name] = {}
+            nested_dict[study_name][trial_id] = data_dict
+        return nested_dict
+
+    def load_params(self):
+        trial_id_col = 'trial_id'
+        param_name_col = 'param_name'
+        param_value_col = 'param_value'
+        distribution_json_col = 'distribution_json'
+        nested_dict = {}
+        for _, row in self.trial_params.iterrows():
+            trial_id = row[trial_id_col]
+            param_name = row[param_name_col]
+            param_value = row[param_value_col]
+            distribution_json = row[distribution_json_col]
+
+            if trial_id not in nested_dict:
+                nested_dict[trial_id] = {}
+
+            dist_json = json.loads(distribution_json)
+            default_step = None
+            default_low = None
+            default_high = None
+            default_log = None
+
+            nested_dict[trial_id][param_name] = {
+                'param_name': param_name,
+                'param_value': param_value,
+                'step': dist_json["attributes"].get("step", default_step),
+                'low': dist_json["attributes"].get("low", default_low),
+                'high': dist_json["attributes"].get("high", default_high),
+                'log': dist_json["attributes"].get("log", default_log),
+            }
+        return nested_dict