Python help

# Sort pipeline by custom metric context.custom_metrics = sorted(custom_metrics, key=lambda x: x[1], reverse=True) # sort high to low log.info('Sorted custom metrics...') log.info(context.custom_metrics) # Calculate custom metric percentile context.custom_metric_percent = {} for i, (stock, custom_metric) in enumerate(context.custom_metrics): context.custom_metric_percent[stock] = 1.0-(float(i)/len(context.custom_metrics))

For every backtest there is an AlgorithmResult object created and stored. Recorded variables, transactions, end of day positions, and some other data are stored in this object. The backtest, or AlgorithmResult object, can be referenced from a notebook with the get_backtest method (see https://www.quantopian.com/docs/api-reference/research-api-reference#quantopian.research.get_backtest). Once one has a reference to this object then the recorded_vars attribute can be accessed (check out https://www.quantopian.com/docs/api-reference/research-api-reference#quantopian.research.AlgorithmResult.recorded_vars). This will contain the values of any recorded variables which can then be plotted.

So, the first step is to get the value at the 75th percentile in a variable. The power of pandas to the rescue. Assuming that custom_metrics is a pandas series. Use the quantile method (https://pandas.pydata.org/pandas-docs/stable/reference/api/pandas.DataFrame.quantile.html). Next 'record' this variable in ones algo code using the record method. Something like this

# First need to calculate the value at 75th percentile  
# Assume 'custom_metrics' is a pandas series  
value_at_75_pct = context.custom_metrics.quantile(.75)

# Then record this value  
record(my_var=value_at_75_pct)

Run the backtest. Copy the backtest ID. Open a notebook. Fetch the AlgorithmResult object using the backtest ID. Like this...

# Put your own backtest ID below  
my_backtest_id = '5d8ba9ba268a4c29d22e265c'  
bt = get_backtest(my_backtest_id)

One can now access any of the backtest attributes. In this case we want recorded_vars . It can then easily be plotted (again with pandas magic) using the plot method.

# The recorded variables can be fetched as a dataframe with the attribute 'recorded_vars'  
# The index will be the date. There is a column for each recorded variable.  
# If the recorded variable is called 'my_var' then the values could be plotted like this  
bt.recorded_vars['my_var'].plot()

That should get you started. Good luck.