Skip to content

Latest commit

 

History

History
196 lines (154 loc) · 5.56 KB

first_gen.md

File metadata and controls

196 lines (154 loc) · 5.56 KB
Raw

Advanced tutorials

Navigation

Custom First Generation

When you use polyga, the default behavior is for the app to randomly generate the first generation of polymers. However, some users may want to manually select the first generation. To do this, there are two methods: loading a file or loading a dataframe.

Loading a file

First, let's set up a general planet and land

import os
import pickle

import pandas as pd

from polyga import polygod as pg
from polyga import utils, selection_schemes

from examples import silly_utils

# Load models for property predictions
model_folder = 'examples'
properties = {
          'Polymer_Coolness': 'Polymer_Coolness_random-forest.pkl',
          'Polymer_Intelligence': 'Polymer_Intelligence_random-forest.pkl',
          'Polymer_Funnyness': 'Polymer_Funnyness_random-forest.pkl'
         }
models = {}

for prop in properties:
    model_file = open(os.path.join(model_folder, properties[prop]), 'rb')
    models[prop] = pickle.load(model_file)
    model_file.close()

planet = pg.PolyPlanet('Planet_Funny', 
        predict_function=silly_utils.silly_property_prediction,
        fingerprint_function=silly_utils.silly_fingerprint,
        models=models
        )

land = pg.PolyLand('Funnyland', planet, 
        generative_function=utils.chromosome_ids_to_smiles,
        fitness_function=silly_utils.make_funniest_polymer
        )

Now, maybe we ran a previous run and found several promising polymers that were super cool and smart, but not very funny. We want to run again with these cool and smart polymers, but bias them towards being funny this time around. For that reason, we've used a new fitness function called "make_funniest_polymer.

We've saved these cool and smart polymers in a folder in our local directory called "cool_smart.csv".

import os
import pandas as pd
import numpy as np
from sklearn.preprocessing import MinMaxScaler

import polyga.analysis as pga
save_loc = 'Planet_Silly'
df, fp_df = pga.load_planet(save_loc)

def find_smart_and_cool(df):
    """Finds smart and cool polymers and saves them to file."""
    properties = ['Polymer_Coolness', 'Polymer_Intelligence']
    df = df.copy()
    fitness = []
    scaled_dict = {}
    for prop in properties:
        vals = np.array(df[prop].values)
        vals = np.reshape(vals, (len(vals), -1))
        scaler = MinMaxScaler()
        scaled_vals = scaler.fit_transform(vals)
        scaled_dict[prop] = scaled_vals

    for i in range(len(df)):
        fit = 0
        for prop in properties:
            fit += 1/3 * scaled_dict[prop][i][0]
        fitness.append(fit)

    df['fitness'] = fitness
    sdf = df.nlargest(180, 'fitness')
    sdf.to_csv('cool_smart.csv', index=False) 

find_smart_and_cool(df)

They must be saved in csv format.

Now pass them to a new nation and run.

nation = pg.PolyNation('FunnyPoly', land,
                       selection_scheme=selection_schemes.elite,
                       partner_selection='diversity', 
                       num_population_initial=180,
                       initial_population_file='cool_smart.csv',
                       )

for i in range(10):
    planet.advance_time()

planet.complete_run()

We could also have loaded the dataframe first and passed that.

df = pd.read_csv('cool_smart.csv')
nation = pg.PolyNation('FunnyPoly', land,
                       selection_scheme=selection_schemes.elite,
                       partner_selection='diversity', 
                       num_population_initial=180,
                       initial_population=df
                       )

for i in range(10):
    planet.advance_time()

planet.complete_run()

We just use initial_population instead of initial_population_file.

These dataframes and files must contain the following columns in order to load properly: 'chromosome_ids', 'num_chromosomes', 'parent_1_id', 'parent_2_id', 'smiles_string', 'birth_land', 'birth_nation', 'birth_planet'.

You can add fillers for these columns, but they must be in the file or dataframe. Chromosome_ids, smiles_string and num_chromosomes must be correct and not fillers though. The full code is below.

import os
import pickle

import pandas as pd

from polyga import polygod as pg
from polyga import utils, selection_schemes

from examples import silly_utils

# Load models for property predictions
model_folder = 'examples'
properties = {
          'Polymer_Coolness': 'Polymer_Coolness_random-forest.pkl',
          'Polymer_Intelligence': 'Polymer_Intelligence_random-forest.pkl',
          'Polymer_Funnyness': 'Polymer_Funnyness_random-forest.pkl'
         }
models = {}

for prop in properties:
    model_file = open(os.path.join(model_folder, properties[prop]), 'rb')
    models[prop] = pickle.load(model_file)
    model_file.close()

planet = pg.PolyPlanet('Planet_Funny', 
        predict_function=silly_utils.silly_property_prediction,
        fingerprint_function=silly_utils.silly_fingerprint,
        models=models
        )

land = pg.PolyLand('Funnyland', planet, 
        generative_function=utils.chromosome_ids_to_smiles,
        fitness_function=silly_utils.make_funniest_polymer
        )
        
nation = pg.PolyNation('FunnyPoly', land,
                       selection_scheme=selection_schemes.elite,
                       partner_selection='diversity', 
                       num_population_initial=180,
                       initial_population_file='cool_smart.csv',
                       )

for i in range(10):
    planet.advance_time()

planet.complete_run()