{ "cells": [ { "cell_type": "markdown", "metadata": {}, "source": [ "# Evolutionary Algorithm" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "the Evolutionary Algorithm that is supposed to create new recipes based on the Recipe Matrices that are created during the *Recipe Analysis* step.\n", "\n", "The Population of the Evolutional Algorithm consists of a set of recipe trees. Each Recipe Tree consists of several Nodes where each node is of one of the following Types:\n", "\n", "* **Ingredient Node:**\n", " these are the leaf nodes. Containing an ingredient. The score is determined by the actions, that are applied if you follow up the path. At the Moment it measures how many duplicate actions are applied.\n", "* **Action Node:**\n", " An Action that is applied on it's child and this child's subtree. Score indicates the average likelihood that this action is applied on the ingredients inside the subtree\n", "* **Mix Node:**\n", " Mixing ingredients together. This is also the only Node that can have more than one child. The score is the average of all pairwise likelihoods that two ingredients are mixed togethter" ] }, { "cell_type": "code", "execution_count": 1, "metadata": {}, "outputs": [], "source": [ "import sys\n", "sys.path.append(\"../\")\n", "sys.path.append(\"../RecipeAnalysis/\")" ] }, { "cell_type": "code", "execution_count": 2, "metadata": {}, "outputs": [ { "data": { "text/html": [ " \n", " " ] }, "metadata": {}, "output_type": "display_data" }, { "data": { "text/html": [ " \n", " " ] }, "metadata": {}, "output_type": "display_data" }, { "name": "stderr", "output_type": "stream", "text": [ "/home/jonas/.local/lib/python3.7/site-packages/ipykernel_launcher.py:39: TqdmExperimentalWarning:\n", "\n", "Using `tqdm.autonotebook.tqdm` in notebook mode. Use `tqdm.tqdm` instead to force console mode (e.g. in jupyter console)\n", "\n" ] } ], "source": [ "import settings\n", "\n", "import pycrfsuite\n", "\n", "import json\n", "\n", "import db.db_settings as db_settings\n", "from db.database_connection import DatabaseConnection\n", "\n", "from Tagging.conllu_generator import ConlluGenerator\n", "from Tagging.crf_data_generator import *\n", "\n", "from RecipeAnalysis.Recipe import Ingredient\n", "\n", "import ea_tools\n", "\n", "from difflib import SequenceMatcher\n", "\n", "import numpy as np\n", "\n", "import ActionGroups as AG\n", "\n", "import plotly.graph_objs as go\n", "from plotly.offline import download_plotlyjs, init_notebook_mode, plot, iplot\n", "from plotly.subplots import make_subplots\n", "init_notebook_mode(connected=True)\n", "\n", "from graphviz import Digraph\n", "\n", "import itertools\n", "\n", "import random\n", "\n", "import plotly.io as pio\n", "pio.renderers.default = \"jupyterlab\"\n", "\n", "from IPython.display import Markdown, HTML, display\n", "\n", "from tqdm.autonotebook import tqdm\n", "\n", "from copy import deepcopy" ] }, { "cell_type": "code", "execution_count": 3, "metadata": {}, "outputs": [], "source": [ "def gaussian(x, mu, sig):\n", " return 1./(np.sqrt(2.*np.pi)*sig)*np.exp(-np.power((x - mu)/sig, 2.)/2)" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "## load adjacency matrices" ] }, { "cell_type": "code", "execution_count": 4, "metadata": {}, "outputs": [], "source": [ "import dill\n", "m_act = dill.load(open(\"../RecipeAnalysis/m_act.dill\", \"rb\"))\n", "m_mix = dill.load(open(\"../RecipeAnalysis/m_mix.dill\", \"rb\"))\n", "m_base_act = dill.load(open(\"../RecipeAnalysis/m_base_act.dill\", \"rb\"))\n", "m_base_mix = dill.load(open(\"../RecipeAnalysis/m_base_mix.dill\", \"rb\"))\n", "\n", "\n", "m_grouped_mix = dill.load(open(\"../RecipeAnalysis/m_grouped_mix_raw.dill\", \"rb\"))\n", "m_grouped_act = dill.load(open(\"../RecipeAnalysis/m_grouped_act_raw.dill\", \"rb\"))\n", "m_grouped_base_act = dill.load(open(\"../RecipeAnalysis/m_grouped_base_act_raw.dill\", \"rb\"))\n", "\n", "\n", "#m_act.apply_threshold(3)\n", "#m_mix.apply_threshold(3)\n", "#m_base_act.apply_threshold(5)\n", "#m_base_mix.apply_threshold(5)\n", "\n", "\n", "#c_act = m_act.get_csr()\n", "#c_mix = m_mix.get_csr()\n", "#c_base_act = m_base_act.get_csr()\n", "#c_base_mix = m_base_mix.get_csr()\n", "\n", "m_act.compile()\n", "m_mix.compile()\n", "m_base_act.compile()\n", "m_base_mix.compile()\n", "\n", "m_grouped_mix.compile()\n", "m_grouped_act.compile()\n", "m_grouped_base_act.compile()\n", "\n", "c_act = m_act._csr\n", "c_mix = m_mix._csr\n", "c_base_act = m_base_act._csr\n", "c_base_mix = m_base_mix._csr\n" ] }, { "cell_type": "code", "execution_count": 5, "metadata": {}, "outputs": [], "source": [ "actions = m_act.get_labels()[0]" ] }, { "cell_type": "code", "execution_count": 6, "metadata": {}, "outputs": [], "source": [ "base_ingredients = m_base_mix.get_labels()" ] }, { "cell_type": "code", "execution_count": 7, "metadata": {}, "outputs": [], "source": [ "sym_label_buffer = {}\n", "fw_label_buffer = {}\n", "bw_label_buffer = {}" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "### helper functions for adjacency matrices" ] }, { "cell_type": "code", "execution_count": 8, "metadata": {}, "outputs": [], "source": [ "def get_sym_adjacent(key, m, c):\n", " index = m._label_index[key]\n", " i1 = c[index,:].nonzero()[1]\n", " i2 = c[:,index].nonzero()[0]\n", " \n", " i = np.concatenate((i1,i2))\n", " \n", " if m in sym_label_buffer:\n", " names = sym_label_buffer[m][i]\n", " else:\n", " names = np.array(m.get_labels())\n", " sym_label_buffer[m] = names\n", " names = names[i]\n", " \n", " counts = np.concatenate((c[index, i1].toarray().flatten(), c[i2, index].toarray().flatten()))\n", " \n", " s = np.argsort(-counts)\n", " \n", " return names[s], counts[s]" ] }, { "cell_type": "code", "execution_count": 9, "metadata": {}, "outputs": [], "source": [ "def get_forward_adjacent(key, m, c):\n", " index = m._x_label_index[key]\n", " i = c[index,:].nonzero()[1]\n", " \n", " if m in fw_label_buffer:\n", " names = fw_label_buffer[m][i]\n", " else:\n", " names = np.array(m._y_labels)\n", " fw_label_buffer[m] = names\n", " names = names[i]\n", " \n", " \n", " counts = c[index, i].toarray().flatten()\n", " \n", " s = np.argsort(-counts)\n", " \n", " return names[s], counts[s]" ] }, { "cell_type": "code", "execution_count": 10, "metadata": {}, "outputs": [], "source": [ "def get_backward_adjacent(key, m, c):\n", " index = m._y_label_index[key]\n", " i = c[:,index].nonzero()[0]\n", " \n", " if m in bw_label_buffer:\n", " names = bw_label_buffer[m][i]\n", " else:\n", " names = np.array(m._x_labels)\n", " bw_label_buffer[m] = names\n", " names = names[i]\n", " \n", " \n", " counts = c[i, index].toarray().flatten()\n", " \n", " s = np.argsort(-counts)\n", " \n", " return names[s], counts[s]" ] }, { "cell_type": "code", "execution_count": 11, "metadata": {}, "outputs": [], "source": [ "def sym_sum(key, m, c):\n", " return np.sum(get_sym_adjacent(key,m,c)[1])\n", "\n", "def fw_sum(key, m, c):\n", " return np.sum(get_forward_adjacent(key,m,c)[1])\n", "\n", "def bw_sum(key, m, c):\n", " return np.sum(get_backward_adjacent(key,m,c)[1])" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "#### different score functions" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "##### normalizations" ] }, { "cell_type": "code", "execution_count": 12, "metadata": {}, "outputs": [], "source": [ "def fw_normalization_factor(key, m, c, quotient_func):\n", " ia = m._x_label_index[key]\n", " \n", " occurances = c[ia,:].nonzero()[1]\n", " \n", " return 1. / quotient_func(c[ia,occurances].toarray())\n", "\n", "def bw_normalization_factor(key, m, c, quotient_func):\n", " ib = m._y_label_index[key]\n", " \n", " occurances = c[:,ib].nonzero()[0]\n", " \n", " return 1. / quotient_func(c[occurances,ib].toarray())\n", "\n", "def sym_normalization_factor(key, m, c, quotient_func):\n", " ii = m._label_index[key]\n", " \n", " fw_occurances = c[ii,:].nonzero()[1]\n", " bw_occurances = c[:,ii].nonzero()[0]\n", " \n", " return 1. / quotient_func(np.concatenate(\n", " [c[ii,fw_occurances].toarray().flatten(),\n", " c[bw_occurances,ii].toarray().flatten()]\n", " ))" ] }, { "cell_type": "code", "execution_count": 13, "metadata": {}, "outputs": [], "source": [ "def sym_p_a_given_b(key_a, key_b, m, c, quot_func = np.max):\n", " ia = m._label_index[key_a]\n", " ib = m._label_index[key_b]\n", " \n", " v = c[ia,ib] + c[ib,ia]\n", " \n", " return v * sym_normalization_factor(key_b, m, c, quot_func)\n", "\n", "def fw_p_a_given_b(key_a, key_b, m, c, quot_func = np.max):\n", " ia = m._x_label_index[key_a]\n", " ib = m._y_label_index[key_b]\n", " \n", " v = c[ia,ib]\n", " \n", " return v * bw_normalization_factor(key_b, m, c, quot_func)\n", "\n", "def bw_p_a_given_b(key_a, key_b, m, c, quot_func = np.max):\n", " ia = m._y_label_index[key_a]\n", " ib = m._x_label_index[key_b]\n", " \n", " v = c[ib,ia]\n", " \n", " return v * fw_normalization_factor(key_b, m, c, quot_func)\n" ] }, { "cell_type": "code", "execution_count": 14, "metadata": {}, "outputs": [], "source": [ "def sym_score(key_a, key_b, m, c):\n", "\n", " ia = m._label_index[key_a]\n", " ib = m._label_index[key_b]\n", " \n", " v = c[ia,ib] + c[ib,ia]\n", " \n", " if v == 0:\n", " return 0\n", " \n", " return max((v/sym_sum(key_a, m, c)), (v/sym_sum(key_b, m, c)))\n", "\n", "def asym_score(key_a, key_b, m, c):\n", " ia = m._x_label_index[key_a]\n", " ib = m._y_label_index[key_b]\n", " \n", " v = c[ia,ib]\n", " \n", " if v == 0:\n", " return 0\n", " \n", " return max(v/fw_sum(key_a, m, c), v/bw_sum(key_b, m, c))" ] }, { "cell_type": "code", "execution_count": 15, "metadata": {}, "outputs": [], "source": [ "def p_ingredient_unprepared(base_ing):\n", " ing = Ingredient(base_ing)\n", " base_sum = sym_sum(base_ing, m_base_mix, c_base_mix)\n", " specialized_sum = sym_sum(ing.to_json(), m_mix, c_mix)\n", " return specialized_sum / base_sum" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "**new probability for preprocess ingredients:**" ] }, { "cell_type": "code", "execution_count": 16, "metadata": {}, "outputs": [], "source": [ "def prepare_ratio(ing:str):\n", " keys, values = m_grouped_act.get_backward_adjacent(Ingredient(ing).to_json())\n", " action_dict = dict(zip(keys,values))\n", " return action_dict['prepare'] / action_dict['heat']\n", "\n", "def random_prepare(ing:str):\n", " \"\"\"\n", " returns randomly a boolean value if ing should be prepared, w.r.t. the prepare_ration function\n", " \"\"\"\n", " \n", " return prepare_ratio(ing) > np.random.normal(0.35,0.1)\n", "\n", "def random_heated(ingredient:str):\n", " action_set, action_weights = m_grouped_base_act.get_backward_adjacent(ingredient)\n", " d = dict(zip(action_set, action_weights))\n", " ratio = 1 - d['prepare'] / d['heat']\n", " \n", " return ratio > np.random.normal(0.65,0.15)\n" ] }, { "cell_type": "code", "execution_count": 17, "metadata": {}, "outputs": [], "source": [ "def relative_action_rank(ingredient:str, action:str):\n", " action_set, action_weights = m_base_act.get_backward_adjacent(ingredient)\n", " if action not in action_set or len(action_set) <= 1:\n", " return 0\n", " return 1 - action_set.tolist().index(action) / (len(action_set) - 1)" ] }, { "cell_type": "code", "execution_count": 18, "metadata": {}, "outputs": [], "source": [ "def filter_set_by_group(act_set, act_w, group):\n", " new_act_set = []\n", " new_act_w = []\n", " for i in range(len(act_set)):\n", " if act_set[i] in AG.inverse_groups[group]:\n", " new_act_set.append(act_set[i])\n", " new_act_w.append(act_w[i])\n", " return np.array(new_act_set), np.array(new_act_w)" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "## better normalized scores:" ] }, { "cell_type": "code", "execution_count": 31, "metadata": {}, "outputs": [], "source": [ "def normalized_score(key, matrix):\n", " sum_key = matrix.get_sum(key)\n", " keys, values = matrix.get_adjacent(key)\n", " normalized_values = np.array([(values[i] / matrix.get_sum(keys[i])) * (values[i] / sum_key) for i in range(len(keys))])\n", " sort = np.argsort(-normalized_values)\n", " return keys[sort], normalized_values[sort]\n", "\n", "def forward_normalized_score(key, matrix):\n", " sum_key = matrix.get_fw_sum(key)\n", " keys, values = matrix.get_forward_adjacent(key)\n", " normalized_values = np.array([(values[i] / matrix.get_bw_sum(keys[i])) * (values[i] / sum_key) for i in range(len(keys))])\n", " sort = np.argsort(-normalized_values)\n", " return keys[sort], normalized_values[sort]\n", "\n", "def backward_normalized_score(key, matrix):\n", " sum_key = matrix.get_bw_sum(key)\n", " keys, values = matrix.get_backward_adjacent(key)\n", " normalized_values = np.array([(values[i] / matrix.get_fw_sum(keys[i])) * (values[i] / sum_key) for i in range(len(keys))])\n", " sort = np.argsort(-normalized_values)\n", " return keys[sort], normalized_values[sort]" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "## Recipe Tree\n", "### Tree Node Base Class" ] }, { "cell_type": "code", "execution_count": 20, "metadata": {}, "outputs": [], "source": [ "class RecipeTreeNode(object):\n", " \n", " id = 0\n", " \n", " def __init__(self, name, constant=False, single_child=False):\n", " self._constant = constant\n", " self._name = name\n", " self._parent = None\n", " \n", " self._id = str(RecipeTreeNode.id)\n", " RecipeTreeNode.id += 1\n", " \n", " self._single_child = single_child\n", " \n", " if self._single_child:\n", " self._child = None\n", " \n", " def child():\n", " return self._child\n", " \n", " def remove_child(c):\n", " assert c == self._child\n", " self._child._parent = None\n", " self._child = None\n", " \n", " def childs():\n", " c = self.child()\n", " if c is None:\n", " return set()\n", " return set([c])\n", " \n", " def add_child(n):\n", " self._child = n\n", " n._parent = self\n", " \n", " self.child = child\n", " self.childs = childs\n", " self.add_child = add_child\n", " self.remove_child = remove_child\n", " else:\n", " self._childs = set()\n", " \n", " def childs():\n", " return self._childs\n", " \n", " def add_child(n):\n", " self._childs.add(n)\n", " n._parent = self\n", " \n", " def remove_child(c):\n", " assert c in self._childs\n", " c._parent = None\n", " self._childs.remove(c)\n", " \n", " self.childs = childs\n", " self.add_child = add_child\n", " self.remove_child = remove_child\n", " \n", " def parent(self):\n", " return self._parent\n", " \n", " def root(self):\n", " if self._parent is None:\n", " return self\n", " return self._parent.root()\n", " \n", " def name(self):\n", " return self._name\n", " \n", " def traverse(self):\n", " l = []\n", " \n", " for c in self.childs():\n", " l += c.traverse()\n", " \n", " return [self] + l\n", " \n", " def traverse_ingredients(self):\n", " ingredient_set = []\n", " for c in self.childs():\n", " ingredient_set += c.traverse_ingredients()\n", " \n", " return ingredient_set\n", " \n", " def remove(self):\n", " p = self.parent()\n", " childs = self.childs().copy()\n", " \n", " assert p is None or not (len(childs) > 1 and p._single_child)\n", " \n", " for c in childs:\n", " self.remove_child(c)\n", " \n", " if p is not None:\n", " p.remove_child(self)\n", " \n", " if self._single_child and self._child is not None and p._name == self._child._name:\n", " # two adjacent nodes with same name would remain after deletion.\n", " # merge them! (by adding the child's childs to our parent instead of our childs)\n", " childs = self._child.childs()\n", " self._child.remove()\n", " \n", " \n", " for c in childs:\n", " p.add_child(c)\n", " \n", " def insert_before(self, n):\n", " p = self._parent\n", " if p is not None:\n", " p.remove_child(self)\n", " p.add_child(n)\n", " n.add_child(self)\n", " \n", " def mutate(self):\n", " n_node = self.n_node_mutate_options()\n", " n_edge = self.n_edge_mutate_options()\n", " \n", " choice = random.choice(range(n_node + n_edge))\n", " if choice < n_node:\n", " self.mutate_node()\n", " else:\n", " self.mutate_edges()\n", " \n", " def mutate_edges(self):\n", " ings = self.traverse_ingredients()\n", " ing = random.choice(ings)\n", " \n", " a, w = get_backward_adjacent(ing._base_ingredient, m_base_act, c_base_act)\n", " \n", " action = random.choices(a, w)[0]\n", " self.insert_before(ActionNode(action))\n", " \n", " def mutate_node(self):\n", " raise NotImplementedError\n", " \n", " def n_node_mutate_options(self):\n", " \n", " return 0 if self._constant else 1\n", " \n", " def n_edge_mutate_options(self):\n", " n = 1 if self._parent is not None else 0\n", " return n\n", " \n", " def n_mutate_options(self):\n", " return self.n_edge_mutate_options() + self.n_node_mutate_options()\n", " \n", " def dot_node(self, dot):\n", " raise NotImplementedError()\n", " \n", " def dot(self, d=None):\n", " if d is None:\n", " d = Digraph()\n", " self.dot_node(d)\n", " \n", " else:\n", " self.dot_node(d)\n", " if self._parent is not None:\n", " d.edge(self._parent._id, self._id)\n", " \n", " \n", " for c in self.childs():\n", " c.dot(d)\n", " \n", " return d\n", " \n", " def serialize(self):\n", " r = {}\n", " r['type'] = str(self.__class__.__name__)\n", " r['id'] = self._id\n", " r['parent'] = self._parent._id if self._parent is not None else None\n", " r['name'] = self._name\n", " r['childs'] = [c._id for c in self.childs()]\n", " r['constant'] = self._constant\n", " r['single_child'] = self._single_child\n", " \n", " return r\n", " \n", " def node_score(self):\n", " raise NotImplementedError()\n", " \n" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "### Mix Node" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "For the Node Score: just make a simple lookup whether this combination is seen or not. So the node Score is defined as:\n" ] }, { "cell_type": "code", "execution_count": 21, "metadata": {}, "outputs": [], "source": [ "class MixNode(RecipeTreeNode):\n", " def __init__(self, constant=False):\n", " super().__init__(\"mix\", constant, single_child=False)\n", " \n", " def dot_node(self, dot):\n", " dot.node(self._id, label=f\"< {self._name}
node score: {self.node_score():.4f}>\", shape=\"diamond\", style=\"filled\", color=\"#d5e8d4\")\n", " \n", " def split(self, set_above, set_below, node_between):\n", " assert len(set_above.difference(self.childs())) == 0\n", " assert len(set_below.difference(self.childs())) == 0\n", " \n", " n_above = MixNode()\n", " n_below = MixNode()\n", " \n", " p = self.parent()\n", " \n", " for c in self.childs().copy():\n", " self.remove_child(c)\n", " self.remove()\n", " \n", " for c in set_below:\n", " n_below.add_child(c)\n", " \n", " for c in set_above:\n", " n_above.add_child(c)\n", " \n", " n_above.add_child(node_between)\n", " node_between.add_child(n_below)\n", " \n", " if p is not None:\n", " p.add_child(n_above)\n", " \n", " # test whether the mix nodes are useless\n", " if len(n_above.childs()) == 1:\n", " n_above.remove()\n", " \n", " if len(n_below.childs()) == 1:\n", " n_below.remove()\n", " \n", " def n_node_mutate_options(self):\n", " return 0 if self._constant or len(self.childs()) <= 2 else len(self.childs())\n", " \n", " def mutate_node(self):\n", " \n", " childs = self.childs()\n", " \n", " if len(childs) <= 2:\n", " print(\"Warning: cannot modify mix node\")\n", " return\n", " \n", " childs = random.sample(childs, len(childs))\n", " \n", " n = random.choice(range(1, len(childs)-1))\n", " \n", " between_node = ActionNode(random.choice(actions))\n", " \n", " self.split(set(childs[:n]), set(childs[n:]), between_node)\n", " \n", " \n", " def node_score(self):\n", " child_ingredients = [c.traverse_ingredients() for c in self.childs()]\n", " \n", " tmp_set = set()\n", " cumulative_sets = []\n", " \n", " pairwise_tuples = []\n", " \n", " for c in child_ingredients:\n", " if len(tmp_set) > 0:\n", " cumulative_sets.append(tmp_set)\n", " pairwise_tuples += [x for x in itertools.product(tmp_set, c)]\n", " tmp_set = tmp_set.union(set(c))\n", " \n", " s_base = 0\n", " s = 0\n", " \n", " for ing_a, ing_b in pairwise_tuples:\n", " try:\n", " #s_base += sym_score(ing_a._base_ingredient, ing_b._base_ingredient, m_base_mix, c_base_mix)\n", " \n", " #s += sym_score(ing_a.to_json(), ing_b.to_json(), m_mix, c_mix)\n", " \n", " # old method:\n", " #p1 = sym_p_a_given_b(ing_a.to_json(), ing_b.to_json(), m_mix, c_mix)\n", " #p2 = sym_p_a_given_b(ing_b.to_json(), ing_a.to_json(), m_mix, c_mix)\n", " #s += 0.5 * p1 + 0.5 * p2\n", " \n", " \n", " ia = m_mix._label_index[ing_a.to_json()]\n", " ib = m_mix._label_index[ing_b.to_json()]\n", " \n", " if c_mix[ia,ib] > 0 or c_mix[ib,ia] > 0:\n", " s += 1\n", " \n", " \n", " \n", " except KeyError as e:\n", " pass\n", " \n", " #s_base /= len(pairwise_tuples)\n", " s /= len(pairwise_tuples)\n", " \n", " #return 0.5 * (s_base + s)\n", " return s\n", " \n", " \n", " \n" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "### Ingredient Node Class" ] }, { "cell_type": "code", "execution_count": 22, "metadata": {}, "outputs": [], "source": [ "class IngredientNode(RecipeTreeNode):\n", " def __init__(self, name, constant=False):\n", " super().__init__(name, constant, single_child=True)\n", " \n", " def get_actions(self):\n", " a_list = []\n", " n = self.parent()\n", " while n is not None:\n", " if type(n) == ActionNode:\n", " a_list.append(n.name())\n", " n = n.parent()\n", " return a_list\n", " \n", " def mutate_node(self):\n", " self._name = random.choice(base_ingredients)\n", " #TODO: change w.r.t. mixing probabilities \n", " \n", " def traverse_ingredients(self):\n", " return [Ingredient(self._name)]\n", " \n", " def node_score(self):\n", " actions = self.get_actions()\n", " \n", " if len(actions) == 0:\n", " if p_ingredient_unprepared(self._name) < 0.2:\n", " return 0\n", " return 1\n", " \n", " seen_actions = set()\n", " n_duplicates = 0\n", " for act in actions:\n", " if act in seen_actions:\n", " n_duplicates += 1\n", " else:\n", " seen_actions.add(act)\n", " \n", " duplicate_actions_score = len(seen_actions) / len(actions)\n", " \n", " return duplicate_actions_score\n", " \n", " \n", " def dot_node(self, dot):\n", " dot.node(self._id, label=f\"< {self._name}
node score:{self.node_score():.4f}>\", shape=\"box\", style=\"filled\", color=\"#ffe6cc\")" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "### Action Node Class" ] }, { "cell_type": "code", "execution_count": 23, "metadata": {}, "outputs": [], "source": [ "class ActionNode(RecipeTreeNode):\n", " def __init__(self, name, constant=False):\n", " super().__init__(name, constant, single_child=True)\n", " \n", " def n_node_mutate_options(self):\n", " # beacause we can change or remove ourselve!\n", " return 0 if self._constant else 2 \n", " def mutate_node(self):\n", " if random.choice(range(2)) == 0:\n", " # change action\n", " self._name = random.choice(actions)\n", " else:\n", " # delete\n", " self.remove()\n", " \n", " def traverse_ingredients(self):\n", " ingredient_set = super().traverse_ingredients()\n", " for ing in ingredient_set:\n", " ing.apply_action(self._name)\n", " \n", " return ingredient_set\n", " \n", " def node_score(self):\n", " ings = self.child().traverse_ingredients()\n", " \n", " s = 0\n", " \n", " for ing in ings:\n", " try:\n", " #score = asym_score(self._name, ing.to_json(), m_act, c_act)\n", " #base_score = asym_score(self._name, ing._base_ingredient, m_base_act, c_base_act)\n", " \n", " score = fw_p_a_given_b(self._name, ing._base_ingredient, m_base_act, c_base_act)\n", " \n", " s += score\n", " except KeyError as e:\n", " pass\n", " \n", " \n", " return s / len(ings)\n", " \n", " def dot_node(self, dot):\n", " dot.node(self._id, label=f\"< {self._name}
node score: {self.node_score():.4f}>\", shape=\"ellipse\", style=\"filled\", color=\"#dae8fc\")" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "### Tree Class" ] }, { "cell_type": "code", "execution_count": 24, "metadata": {}, "outputs": [], "source": [ "class Tree(object):\n", " @staticmethod\n", " def build_initial_tree(ingredients: list, main_ingredients: list, max_n = 4, wheel_turns = 2):\n", " \n", " '''\n", " # get action sets for ingredients\n", " possible_actions = {}\n", " for ing in ingredients:\n", " action_set, action_weights = m_base_act.get_backward_adjacent(ing)\n", " possible_actions[ing] = set(action_set.tolist()[:5])\n", " \n", " # now find actions with the same subset\n", " \n", " ings_for_acts = {}\n", " \n", " for ing, acts in possible_actions.items():\n", " for a in acts:\n", " if a not in ings_for_acts:\n", " ings_for_acts[a] = set()\n", " \n", " ings_for_acts[a].add(ing)\n", "\n", " '''\n", " \n", " # choose randomly an action for each ingredient by the \"wheel of fortune\" method\n", " actions_for_ing = {}\n", " for ing in ingredients:\n", " actions_for_ing[ing] = set()\n", " action_set, action_weights = m_base_act.get_backward_adjacent(ing)\n", " for i in range(wheel_turns):\n", " action = ea_tools.wheel_of_fortune_selection(action_set[:max_n], action_weights[:max_n])\n", " actions_for_ing[ing].add(action)\n", " #print(ing, action)\n", "\n", " ings_for_acts = {}\n", " \n", " for ing, acts in actions_for_ing.items():\n", " for a in acts:\n", " if a not in ings_for_acts:\n", " ings_for_acts[a] = set()\n", " \n", " ings_for_acts[a].add(ing)\n", " \n", " # now looking for the largest subset and choose one of them randomly\n", " \n", " action_keys = np.array(list(ings_for_acts.keys()))\n", " set_lengths = np.array([len(ings_for_acts[a]) for a in action_keys])\n", " \n", " # sort lengths\n", " sorted_length_indices = np.argsort(-set_lengths)\n", " \n", " # now perform the following steps:\n", " # * go through all unprocessed ingredients\n", " # * for each ing: find largest action set that is suitable\n", " # * perform this action on all it's ingredients.\n", " # * continue until no ingredient is left\n", " \n", " unprocessed_ings = set(ingredients)\n", " unprocessed_actions = set(ings_for_acts.keys())\n", " \n", " ingredient_nodes = {}\n", " \n", " # create ingredient nodes:\n", " for ing in ingredients:\n", " ingredient_nodes[ing] = IngredientNode(ing, constant=True)\n", " \n", " i = 0\n", " \n", " while len(unprocessed_ings) > 0:\n", " \n", " # select random ingredient:\n", " ing = np.random.choice(list(unprocessed_ings))\n", " \n", " sorted_actions = action_keys[sorted_length_indices]\n", " selected_action = None\n", " \n", " for action in sorted_actions:\n", " if ing in ings_for_acts[action]:\n", " selected_action = action\n", " break\n", " \n", " # found best action. apply it to all matching ingredients\n", " if selected_action is not None:\n", " matching_ingredients = ings_for_acts[selected_action]\n", " \n", " # debugging:\n", " '''\n", " print(f\"choose {selected_action}\")\n", " print(f\"matching ingredients {matching_ingredients}\")\n", " '''\n", " \n", " if len(matching_ingredients) == 1:\n", " ing = list(matching_ingredients)[0]\n", " ing_node = ingredient_nodes[ing].root()\n", " action_node = ActionNode(selected_action)\n", " action_node.add_child(ing_node)\n", " unprocessed_ings.remove(ing)\n", " #display(action_node.dot())\n", " \n", " else:\n", " \n", " nodes_to_mix = set()\n", "\n", " mix_node = MixNode()\n", " action_node = ActionNode(selected_action)\n", " action_node.add_child(mix_node)\n", " \n", " for ing in matching_ingredients:\n", " nodes_to_mix.add(ingredient_nodes[ing].root())\n", " \n", " if ing in unprocessed_ings:\n", " unprocessed_ings.remove(ing)\n", "\n", " for node in nodes_to_mix:\n", " mix_node.add_child(node)\n", " #display(action_node.dot())\n", " \n", " # debugging:\n", " '''\n", " tmp = set([n.root() for n in ingredient_nodes.values()])\n", " print(f\"iteration {i}:\")\n", " for n in tmp:\n", " print(n.name())\n", " display(n.dot())\n", " '''\n", " i += 1\n", " \n", " \n", " root_layer = set([n.root() for n in ingredient_nodes.values()])\n", "\n", " root_layer_without_parents = []\n", " for node in root_layer:\n", " if node.parent() is None:\n", " root_layer_without_parents.append(node)\n", " \n", " if len(root_layer_without_parents) == 1:\n", " return root_layer_without_parents[0]\n", " \n", " root_node = MixNode()\n", " for r in root_layer_without_parents:\n", " root_node.add_child(r)\n", " \n", " return root_node\n", " \n", " @staticmethod\n", " def find_ingredients(constant_ingredients, main_ingredients, min_additional:int, max_additional:int, top_ings:int=3):\n", " '''\n", " create an initial set of ingredients, based on given constant ingredients.\n", " min_additional and max_additional gives the range of ingredients that are added to our set\n", " '''\n", " \n", " seen_items = set(constant_ingredients)\n", " \n", " items = []\n", " scores = []\n", " \n", " for ing in constant_ingredients:\n", " # find best matching ingredients\n", " best_items = []\n", " best_scores = []\n", " \n", " candidates, weights = m_base_mix.get_adjacent(ing)\n", " i = 0\n", " while i < len(candidates) and len(best_items) < top_ings:\n", " if candidates[i] not in seen_items:\n", " best_items.append(candidates[i])\n", " best_scores.append(weights[i])\n", " i += 1\n", " \n", " items.append(best_items)\n", " scores.append(best_scores)\n", " \n", " #TODO: error handling if too few options are availabale!\n", " \n", " additional_ingredients = ea_tools.combined_wheel_of_fortune_selection(items,\n", " scores,\n", " np.random.randint(\n", " min_additional,\n", " max_additional + 1\n", " ))\n", " \n", " return list(constant_ingredients) + list(additional_ingredients)\n", "\n", " @staticmethod\n", " def from_ingredients(ingredients: list, main_ingredients: list, additional_ings=0):\n", " root = None\n", " \n", " constant_ingredients = ingredients\n", " \n", " if additional_ings > 0:\n", " ingredients = Tree.find_ingredients(ingredients, main_ingredients, min_additional=0, max_additional=additional_ings)\n", " \n", " \n", " root = Tree.build_initial_tree(ingredients, main_ingredients)\n", " \n", " # mark initial ingredient nodes as constant:\n", " nodes = root.traverse()\n", " for node in nodes:\n", " if type(node) == IngredientNode:\n", " if node.name() in constant_ingredients:\n", " node._constant = True\n", " \n", " return Tree(root)\n", " \n", " @staticmethod\n", " def from_serialization(s):\n", " def empty_node(raw_n):\n", " if raw_n['type'] == \"MixNode\":\n", " node = MixNode(raw_n['constant'])\n", " elif raw_n['type'] == \"IngredientNode\":\n", " node = IngredientNode(raw_n['name'], raw_n['constant'])\n", " elif raw_n['type'] == \"ActionNode\":\n", " node = ActionNode(raw_n['name'], raw_n['constant'])\n", " else:\n", " print(\"unknown node detected\")\n", " return\n", " \n", " return node\n", " \n", " nodes = {}\n", " for n in s:\n", " nodes[n['id']] = empty_node(n)\n", " \n", " for n in s:\n", " childs = n['childs']\n", " id = n['id']\n", " for c in childs:\n", " nodes[id].add_child(nodes[c])\n", " \n", " return Tree(nodes[s[0]['id']])\n", " \n", " \n", " def __init__(self, root):\n", " # create a dummy entry node\n", " self._root = RecipeTreeNode(\"root\", single_child=True)\n", " self._root.add_child(root)\n", " \n", " def root(self):\n", " return self._root.child()\n", " \n", " def mutate(self):\n", " nodes = self.root().traverse()\n", " weights = [n.n_mutate_options() for n in nodes]\n", " \n", " n = random.choices(nodes, weights)[0]\n", " \n", " n.mutate()\n", " \n", " def dot(self):\n", " return self.root().dot()\n", " \n", " def serialize(self):\n", " return [n.serialize() for n in self.root().traverse()]\n", " \n", " def structure_score(self):\n", " n_duplicates = 0\n", " \n", " \n", " def collect_scores(self):\n", " self._mix_scores = []\n", " self._act_scores = []\n", " self._ing_scores = []\n", " \n", " nodes = self.root().traverse()\n", " self._n_mix_nodes = 0\n", " self._n_act_nodes = 0\n", " self._n_ing_nodes = 0\n", " \n", " s = 0\n", " for n in nodes:\n", " if type(n) == MixNode:\n", " self._mix_scores.append(n.node_score())\n", " self._n_mix_nodes += 1\n", " if type(n) == ActionNode:\n", " self._act_scores.append(n.node_score())\n", " self._n_act_nodes += 1\n", " if type(n) == IngredientNode:\n", " self._ing_scores.append(n.node_score())\n", " self._n_ing_nodes += 1\n", " \n", " self._n_duplicates = 0\n", " seen_actions = set()\n", " \n", " for n in nodes:\n", " if type(n) == ActionNode:\n", " if n.name() in seen_actions:\n", " self._n_duplicates += 1\n", " else:\n", " seen_actions.add(n.name())\n", " \n", " self._mix_scores = np.array(self._mix_scores)\n", " self._act_scores = np.array(self._act_scores)\n", " self._ing_scores = np.array(self._ing_scores)\n", " \n", " \n", " def mix_scores(self):\n", " return self._mix_scores\n", " \n", " def action_scores(self):\n", " return self._act_scores\n", " \n", " def ing_scores(self):\n", " return self._ing_scores\n", " \n", " def bounds_mix_scores(self):\n", " \n", " nonzeros = self._mix_scores[self._mix_scores > 0]\n", " if len(nonzeros) > 0:\n", " mmax = np.max(nonzeros)\n", " mmin = np.min(nonzeros)\n", " return mmin, mmax\n", " else:\n", " return None, None\n", " \n", " def bounds_act_scores(self):\n", " \n", " if len(self._act_scores) == 0:\n", " return None, None\n", " nonzeros = self._act_scores[self._act_scores > 0]\n", " if len(nonzeros) > 0:\n", " mmax = np.max(nonzeros)\n", " mmin = np.min(nonzeros)\n", " return mmin, mmax\n", " else:\n", " return None, None\n", " \n", " def normalized_mix_scores(self, min, max):\n", " if (max != min):\n", " normalized = (self._mix_scores - min)/(max-min)\n", " normalized[normalized <= 0] = 0\n", " return normalized\n", " else:\n", " return None\n", " \n", " def normalized_act_scores(self, min, max):\n", " if len(self._act_scores) == 0 or max == min:\n", " return None\n", " normalized = (self._act_scores - min)/(max-min)\n", " normalized[normalized <= 0] = 0\n", " return normalized\n", " \n", " def copy(self):\n", " return Tree.from_serialization(self.serialize())\n" ] }, { "cell_type": "code", "execution_count": 25, "metadata": {}, "outputs": [], "source": [ "ingredients = [\"onion\", \"tomato\", \"rice\", \"salt\"] \n", "main_ingredients = [\"rice\"]\n", "\n", "assert set(main_ingredients).issubset(set(ingredients))\n", "\n", "max_n = 5\n", "wheel_turns = 2\n", "\n", "def does_action_match(ingredient:str, action:str, t = 0.6):\n", " return relative_action_rank(ingredient, action) > t\n", "\n", "\n", "# choose randomly an action for each ingredient by the \"wheel of fortune\" method\n", "actions_for_ing = {}\n", "for ing in ingredients:\n", " actions_for_ing[ing] = set()\n", " action_set, action_weights = m_base_act.get_backward_adjacent(ing)\n", " if random_heated(ing):\n", " #print(action_set)\n", " action_set, action_weights = filter_set_by_group(action_set, action_weights, \"heat\")\n", " #print(action_set)\n", " for i in range(wheel_turns):\n", " if ing in main_ingredients:\n", " # if main ingredient: choose by action probability\n", " w = np.array(list(action_weights), dtype=float)\n", " w *= (1.0 / np.sum(w))\n", " action = np.random.choice(list(action_set), size=1, replace=False, p=w)[0]\n", " else:\n", " # else: choose rank based\n", " action = ea_tools.wheel_of_fortune_selection(action_set[:max_n], action_weights[:max_n])\n", " actions_for_ing[ing].add(action)\n", " #print(f\"action {action} for ing {ing}\")\n", " #print(ing, action)\n", " \n", "# create ingredient nodes:\n", "ingredient_nodes = {}\n", "\n", "# create ingredient nodes:\n", "for ing in ingredients:\n", " new_node = IngredientNode(ing, constant=True)\n", " \n", " # check if we should do a preparation step\n", " if random_prepare(ing):\n", " # choose a preparation cooking action\n", " action_set, action_weights = m_act.get_backward_adjacent(Ingredient(ing).to_json())\n", " action_set, action_weights = filter_set_by_group(action_set, action_weights, \"prepare\")\n", " if len(action_set) > 0:\n", " action = ea_tools.wheel_of_fortune_selection(action_set[:max_n], action_weights[:max_n])\n", " act_node = ActionNode(action)\n", " act_node.add_child(new_node)\n", " new_node = act_node\n", " \n", " \n", " ingredient_nodes[ing] = new_node\n", "\n", "# starting now with the actions found for the main ingredients and try to match all ingredients together\n", "# with that:\n", "\n", "unprocessed_ings = set(filter(lambda x: len(actions_for_ing[x]) > 0, ingredients))\n", "unprocessed_main_ings = set(filter(lambda x: len(actions_for_ing[x]) > 0, main_ingredients))\n", "\n", "while len(unprocessed_main_ings) > 0:\n", " main_ing = unprocessed_main_ings.pop()\n", " \n", " # random action for that ing:\n", " act = actions_for_ing[main_ing].pop()\n", " \n", " act_node = ActionNode(act)\n", " mix_node = MixNode()\n", " mix_node.add_child(ingredient_nodes[main_ing])\n", " act_node.add_child(mix_node)\n", " ingredient_nodes[main_ing] = act_node\n", " \n", " unprocessed_ings.remove(main_ing)\n", " \n", " for ing in unprocessed_ings.copy():\n", " if does_action_match(ing, act):\n", " mix_node.add_child(ingredient_nodes[ing])\n", " ingredient_nodes[ing] = act_node\n", " unprocessed_ings.remove(ing)\n", " if ing in unprocessed_main_ings:\n", " unprocessed_main_ings.remove(ing)\n", " \n", " if len(mix_node.childs()) == 1:\n", " mix_node.remove()\n", "\n", "# now make the same with all remaining ingredients:\n", "while len(unprocessed_ings) > 0:\n", " current_ing = unprocessed_ings.pop() \n", " \n", " # random action for that ing:\n", " act = actions_for_ing[current_ing].pop()\n", " \n", " act_node = ActionNode(act)\n", " mix_node = MixNode()\n", " mix_node.add_child(ingredient_nodes[current_ing])\n", " act_node.add_child(mix_node)\n", " \n", " ingredient_nodes[current_ing] = act_node\n", " \n", " \n", " for ing in unprocessed_ings.copy():\n", " if does_action_match(ing, act):\n", " mix_node.add_child(ingredient_nodes[ing])\n", " ingredient_nodes[ing] = act_node\n", " unprocessed_ings.remove(ing)\n", " \n", " if len(mix_node.childs()) == 1:\n", " mix_node.remove()\n", "\n", "\n", "root_layer = set([n.root() for n in ingredient_nodes.values()])\n", "\n", "root_layer_without_parents = []\n", "for node in root_layer:\n", " if node.parent() is None:\n", " root_layer_without_parents.append(node)\n", "\n", "if len(root_layer_without_parents) == 1:\n", " root_node = root_layer_without_parents[0]\n", "\n", "else:\n", " root_node = MixNode()\n", " for r in root_layer_without_parents:\n", " root_node.add_child(r)\n", "\n" ] }, { "cell_type": "code", "execution_count": 26, "metadata": {}, "outputs": [ { "data": { "image/svg+xml": [ "\n", "\n", "\n", "\n", "\n", "\n", "%3\n", "\n", "\n", "\n", "10\n", "\n", " \n", "mix\n", "node score: 0.0000\n", "\n", "\n", "\n", "8\n", "\n", " \n", "bake\n", "node score: 1.0000\n", "\n", "\n", "\n", "10->8\n", "\n", "\n", "\n", "\n", "\n", "3\n", "\n", " \n", "cut\n", "node score: 0.2505\n", "\n", "\n", "\n", "10->3\n", "\n", "\n", "\n", "\n", "\n", "6\n", "\n", " \n", "fry\n", "node score: 0.0859\n", "\n", "\n", "\n", "10->6\n", "\n", "\n", "\n", "\n", "\n", "5\n", "\n", " \n", "salt\n", "node score:1.0000\n", "\n", "\n", "\n", "8->5\n", "\n", "\n", "\n", "\n", "\n", "2\n", "\n", " \n", "tomato\n", "node score:1.0000\n", "\n", "\n", "\n", "3->2\n", "\n", "\n", "\n", "\n", "\n", "7\n", "\n", " \n", "mix\n", "node score: 1.0000\n", "\n", "\n", "\n", "6->7\n", "\n", "\n", "\n", "\n", "\n", "1\n", "\n", " \n", "slice\n", "node score: 0.1501\n", "\n", "\n", "\n", "7->1\n", "\n", "\n", "\n", "\n", "\n", "4\n", "\n", " \n", "rice\n", "node score:1.0000\n", "\n", "\n", "\n", "7->4\n", "\n", "\n", "\n", "\n", "\n", "0\n", "\n", " \n", "onion\n", "node score:1.0000\n", "\n", "\n", "\n", "1->0\n", "\n", "\n", "\n", "\n", "\n" ], "text/plain": [ "" ] }, "execution_count": 26, "metadata": {}, "output_type": "execute_result" } ], "source": [ "root_node.dot()" ] }, { "cell_type": "code", "execution_count": 27, "metadata": {}, "outputs": [], "source": [ "constant_ingredients = [\"noodle\", \"onion\", \"tomato\"]\n", "main_ingredients = ['noodle']\n", "min_additional = 3\n", "max_additional = 6\n", "top_ings =3" ] }, { "cell_type": "code", "execution_count": 36, "metadata": {}, "outputs": [ { "data": { "text/plain": [ "['noodle',\n", " 'onion',\n", " 'tomato',\n", " 'salt',\n", " 'cheese',\n", " 'garlic clove',\n", " 'olive oil',\n", " 'mozzarella cheese']" ] }, "execution_count": 36, "metadata": {}, "output_type": "execute_result" } ], "source": [ "seen_items = set(constant_ingredients)\n", "\n", "items = []\n", "scores = []\n", "\n", "assert set(main_ingredients).issubset(set(constant_ingredients))\n", "\n", "# additional ingredients are choosen w.r.t all given ingredients\n", "n_additional_ings = np.random.randint(min_additional, max_additional + 1)\n", "\n", "# extra ings are ingredients choosen specially for the main ingredient\n", "n_extra_ings = int((len(main_ingredients) / len(constant_ingredients)) * n_additional_ings)\n", "\n", "if n_extra_ings > n_additional_ings:\n", " n_extra_ings = n_additional_ings\n", "\n", " \n", "# choose extra ingredients\n", "extra_candidates = []\n", "extra_weights = []\n", "\n", "for ing in main_ingredients:\n", " candidates, weights = normalized_score(ing, m_base_mix)\n", " extra_candidates.append(candidates[:10])\n", " extra_weights.append(weights[:10])\n", "\n", "extra_ingredients = ea_tools.combined_wheel_of_fortune_selection(extra_candidates,\n", " extra_weights,\n", " n_extra_ings)\n", "\n", "for ing in constant_ingredients:\n", " # find best matching ingredients\n", " best_items = []\n", " best_scores = []\n", "\n", " candidates, weights = m_base_mix.get_adjacent(ing)\n", " i = 0\n", " while i < len(candidates) and len(best_items) < top_ings:\n", " if candidates[i] not in seen_items:\n", " best_items.append(candidates[i])\n", " best_scores.append(weights[i])\n", " i += 1\n", "\n", " items.append(best_items)\n", " scores.append(best_scores)\n", "\n", "#TODO: error handling if too few options are availabale!\n", "\n", "additional_ingredients = ea_tools.combined_wheel_of_fortune_selection(items,\n", " scores,\n", " n_additional_ings - n_extra_ings)\n", "list(constant_ingredients) + list(additional_ingredients) + list(extra_ingredients)" ] }, { "cell_type": "code", "execution_count": 32, "metadata": {}, "outputs": [ { "data": { "text/plain": [ "(array(['ricotta cheese', 'spaghetti sauce', 'mozzarella cheese', 'cheese',\n", " 'spinach', 'sausage', 'ground beef', 'tomato sauce', 'onion',\n", " 'basil', 'broccoli', 'parsley', 'mushroom soup', 'mushroom',\n", " 'sauce', 'egg', 'garlic clove', 'chicken', 'milk', 'water', 'salt',\n", " 'zucchini', 'tomato', 'pepper', 'seasoning', 'green pepper',\n", " 'shrimp', 'soy sauce', 'butter', 'red pepper', 'oregano',\n", " 'clove garlic', 'olive oil', 'pork', 'carrot', 'green onion',\n", " 'cream cheese', 'garlic', 'chicken broth', 'tablespoon butter',\n", " 'red bell pepper', 'flour', 'cream', 'black pepper',\n", " 'vegetable oil', 'chicken breast', 'sugar'], dtype=' self._scores[i_b]:\n", " new_population.append(self.population[i_a])\n", " else:\n", " new_population.append(self.population[i_b])\n", " \n", " self.population = new_population\n", " \n", " def hold_best(self, n=10):\n", " sorted_indices = np.argsort(-self._scores)\n", " \n", " self.population = np.array(self.population)[sorted_indices[:n]].tolist()\n", " \n", " def analyse_scores(self):\n", " for tree in self.population:\n", " min, max = tree.bounds_mix_scores()\n", " if min is not None and max is not None:\n", " if self._mix_min is None or min < self._mix_min:\n", " self._mix_min = min\n", " if self._mix_max is None or max > self._mix_max:\n", " self._mix_max = max\n", " \n", " min, max = tree.bounds_act_scores()\n", " if min is not None and max is not None:\n", " if self._act_min is None or min < self._act_min:\n", " self._act_min = min\n", " if self._act_max is None or max > self._act_max:\n", " self._act_max = max\n", " \n", " def single_score(self, mix_scores, act_scores, ing_scores):\n", " if mix_scores is None or act_scores is None or ing_scores is None:\n", " return 0\n", " score = (0.5 * np.average(mix_scores) + 0.5 * np.average(act_scores)) * np.average(ing_scores)\n", " # judging also how many actions we have. So far use a gaussian with mean at number of ingredients\n", " \n", " score *= gaussian(len(act_scores), len(mix_scores), 1)\n", " return score\n", " \n", " \n", " \n", " def collect_scores(self):\n", " for tree in tqdm(self.population, desc=\"evaluate population scores\", leave=False):\n", " tree.collect_scores()\n", " \n", " self.analyse_scores()\n", " \n", " if self._mix_min is not None and self._mix_max is not None:\n", " self._mix_scores = [t.normalized_mix_scores(self._mix_min, self._mix_max) for t in self.population]\n", " else:\n", " # if no normalization can be done, all values are the same or 0.\n", " # in this case just fill in zeros as score\n", " self._mix_scores = [np.zeros(shape=t._mix_scores.shape) for t in self.population]\n", " \n", " if self._act_min is not None and self._act_max is not None:\n", " self._act_scores = [t.normalized_act_scores(self._act_min, self._act_max) for t in self.population]\n", " else:\n", " self._act_scores = [np.zeros(shape=t._act_scores) for t in self.population]\n", " \n", " self._scores = []\n", " for i in range(len(self._mix_scores)):\n", " #print (self._mix_scores[i], self._act_scores[i])\n", " if self._act_scores is None or self._mix_scores is None or self._act_scores[i] is None:\n", " self._scores.append(0)\n", " continue\n", " \n", " s = self.single_score(self._mix_scores[i], self._act_scores[i], self.population[i].ing_scores())\n", " self._scores.append(s)\n", " self._scores = np.array(self._scores)\n", " \n", " def run(self, n=50):\n", " for i in tqdm(range(n), desc=\"run evolutionary cycles\"):\n", " self.mutate()\n", " self.mutate()\n", " self.collect_scores()\n", " \n", " #self.pairwise_competition()\n", " #self.collect_scores()\n", " self.hold_best(self._n)\n", " \n", " \n", " \n", " def plot_population(self, collect_scores=True):\n", " if (collect_scores):\n", " self.collect_scores()\n", " #print(self._mix_scores)\n", " #print(self._act_scores)\n", " #print(self._scores)\n", " for i, t in enumerate(self.population):\n", " if (collect_scores):\n", " display(self._scores[i])\n", " display(t.root().dot())" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "## Run Evolutionary Algorithm" ] }, { "cell_type": "code", "execution_count": 30, "metadata": {}, "outputs": [], "source": [ "p = Population([\"bacon\", \"tomato\", \"onion\"],['noodle'], max_additional_ings=6)" ] }, { "cell_type": "code", "execution_count": 24, "metadata": {}, "outputs": [], "source": [ "#p_ingredient_unprepared(list(p.population[0].root().childs())[0]._name) < 0.2" ] }, { "cell_type": "code", "execution_count": 26, "metadata": {}, "outputs": [], "source": [ "#p.run(100)" ] }, { "cell_type": "code", "execution_count": 31, "metadata": {}, "outputs": [ { "data": { "image/svg+xml": [ "\n", "\n", "\n", "\n", "\n", "\n", "%3\n", "\n", "\n", "\n", "9\n", "\n", " \n", "mix\n", "node score: 0.1429\n", "\n", "\n", "\n", "5\n", "\n", " \n", "bake\n", "node score: 1.0000\n", "\n", "\n", "\n", "9->5\n", "\n", "\n", "\n", "\n", "\n", "7\n", "\n", " \n", "cook\n", "node score: 0.9182\n", "\n", "\n", "\n", "9->7\n", "\n", "\n", "\n", "\n", "\n", "8\n", "\n", " \n", "beat\n", "node score: 0.7344\n", "\n", "\n", "\n", "9->8\n", "\n", "\n", "\n", "\n", "\n", "3\n", "\n", " \n", "salt\n", "node score:1.0000\n", "\n", "\n", "\n", "5->3\n", "\n", "\n", "\n", "\n", "\n", "6\n", "\n", " \n", "mix\n", "node score: 0.6667\n", "\n", "\n", "\n", "7->6\n", "\n", "\n", "\n", "\n", "\n", "0\n", "\n", " \n", "bacon\n", "node score:1.0000\n", "\n", "\n", "\n", "6->0\n", "\n", "\n", "\n", "\n", "\n", "2\n", "\n", " \n", "onion\n", "node score:1.0000\n", "\n", "\n", "\n", "6->2\n", "\n", "\n", "\n", "\n", "\n", "1\n", "\n", " \n", "tomato\n", "node score:1.0000\n", "\n", "\n", "\n", "6->1\n", "\n", "\n", "\n", "\n", "\n", "4\n", "\n", " \n", "egg\n", "node score:1.0000\n", "\n", "\n", "\n", "8->4\n", "\n", "\n", "\n", "\n", "\n" ], "text/plain": [ "" ] }, "metadata": {}, "output_type": "display_data" }, { "data": { "image/svg+xml": [ "\n", "\n", "\n", "\n", "\n", "\n", "%3\n", "\n", "\n", "\n", "25\n", "\n", " \n", "mix\n", "node score: 0.1000\n", "\n", "\n", "\n", "24\n", "\n", " \n", "heat\n", "node score: 1.0000\n", "\n", "\n", "\n", "25->24\n", "\n", "\n", "\n", "\n", "\n", "20\n", "\n", " \n", "cook\n", "node score: 0.7815\n", "\n", "\n", "\n", "25->20\n", "\n", "\n", "\n", "\n", "\n", "22\n", "\n", " \n", "bake\n", "node score: 1.0000\n", "\n", "\n", "\n", "25->22\n", "\n", "\n", "\n", "\n", "\n", "23\n", "\n", " \n", "mix\n", "node score: 1.0000\n", "\n", "\n", "\n", "24->23\n", "\n", "\n", "\n", "\n", "\n", "14\n", "\n", " \n", "olive oil\n", "node score:1.0000\n", "\n", "\n", "\n", "23->14\n", "\n", "\n", "\n", "\n", "\n", "12\n", "\n", " \n", "tomato\n", "node score:1.0000\n", "\n", "\n", "\n", "23->12\n", "\n", "\n", "\n", "\n", "\n", "19\n", "\n", " \n", "mix\n", "node score: 0.8333\n", "\n", "\n", "\n", "20->19\n", "\n", "\n", "\n", "\n", "\n", "13\n", "\n", " \n", "onion\n", "node score:1.0000\n", "\n", "\n", "\n", "19->13\n", "\n", "\n", "\n", "\n", "\n", "11\n", "\n", " \n", "bacon\n", "node score:1.0000\n", "\n", "\n", "\n", "19->11\n", "\n", "\n", "\n", "\n", "\n", "16\n", "\n", " \n", "cheese\n", "node score:1.0000\n", "\n", "\n", "\n", "19->16\n", "\n", "\n", "\n", "\n", "\n", "17\n", "\n", " \n", "garlic clove\n", "node score:1.0000\n", "\n", "\n", "\n", "19->17\n", "\n", "\n", "\n", "\n", "\n", "21\n", "\n", " \n", "mix\n", "node score: 1.0000\n", "\n", "\n", "\n", "22->21\n", "\n", "\n", "\n", "\n", "\n", "15\n", "\n", " \n", "salt\n", "node score:1.0000\n", "\n", "\n", "\n", "21->15\n", "\n", "\n", "\n", "\n", "\n", "18\n", "\n", " \n", "egg\n", "node score:1.0000\n", "\n", "\n", "\n", "21->18\n", "\n", "\n", "\n", "\n", "\n" ], "text/plain": [ "" ] }, "metadata": {}, "output_type": "display_data" }, { "data": { "image/svg+xml": [ "\n", "\n", "\n", "\n", "\n", "\n", "%3\n", "\n", "\n", "\n", "32\n", "\n", " \n", "heat\n", "node score: 0.8099\n", "\n", "\n", "\n", "31\n", "\n", " \n", "mix\n", "node score: 0.8333\n", "\n", "\n", "\n", "32->31\n", "\n", "\n", "\n", "\n", "\n", "27\n", "\n", " \n", "bacon\n", "node score:1.0000\n", "\n", "\n", "\n", "31->27\n", "\n", "\n", "\n", "\n", "\n", "28\n", "\n", " \n", "tomato\n", "node score:1.0000\n", "\n", "\n", "\n", "31->28\n", "\n", "\n", "\n", "\n", "\n", "29\n", "\n", " \n", "onion\n", "node score:1.0000\n", "\n", "\n", "\n", "31->29\n", "\n", "\n", "\n", "\n", "\n", "30\n", "\n", " \n", "salt\n", "node score:1.0000\n", "\n", "\n", "\n", "31->30\n", "\n", "\n", "\n", "\n", "\n" ], "text/plain": [ "" ] }, "metadata": {}, "output_type": "display_data" }, { "data": { "image/svg+xml": [ "\n", "\n", "\n", "\n", "\n", "\n", "%3\n", "\n", "\n", "\n", "40\n", "\n", " \n", "mix\n", "node score: 0.0000\n", "\n", "\n", "\n", "38\n", "\n", " \n", "heat\n", "node score: 0.9679\n", "\n", "\n", "\n", "40->38\n", "\n", "\n", "\n", "\n", "\n", "39\n", "\n", " \n", "place\n", "node score: 0.3942\n", "\n", "\n", "\n", "40->39\n", "\n", "\n", "\n", "\n", "\n", "37\n", "\n", " \n", "simmer\n", "node score: 0.7959\n", "\n", "\n", "\n", "40->37\n", "\n", "\n", "\n", "\n", "\n", "36\n", "\n", " \n", "onion\n", "node score:1.0000\n", "\n", "\n", "\n", "38->36\n", "\n", "\n", "\n", "\n", "\n", "34\n", "\n", " \n", "bacon\n", "node score:1.0000\n", "\n", "\n", "\n", "39->34\n", "\n", "\n", "\n", "\n", "\n", "35\n", "\n", " \n", "tomato\n", "node score:1.0000\n", "\n", "\n", "\n", "37->35\n", "\n", "\n", "\n", "\n", "\n" ], "text/plain": [ "" ] }, "metadata": {}, "output_type": "display_data" }, { "data": { "image/svg+xml": [ "\n", "\n", "\n", "\n", "\n", "\n", "%3\n", "\n", "\n", "\n", "48\n", "\n", " \n", "mix\n", "node score: 0.0000\n", "\n", "\n", "\n", "46\n", "\n", " \n", "heat\n", "node score: 0.8198\n", "\n", "\n", "\n", "48->46\n", "\n", "\n", "\n", "\n", "\n", "47\n", "\n", " \n", "simmer\n", "node score: 0.7959\n", "\n", "\n", "\n", "48->47\n", "\n", "\n", "\n", "\n", "\n", "45\n", "\n", " \n", "mix\n", "node score: 1.0000\n", "\n", "\n", "\n", "46->45\n", "\n", "\n", "\n", "\n", "\n", "42\n", "\n", " \n", "bacon\n", "node score:1.0000\n", "\n", "\n", "\n", "45->42\n", "\n", "\n", "\n", "\n", "\n", "44\n", "\n", " \n", "onion\n", "node score:1.0000\n", "\n", "\n", "\n", "45->44\n", "\n", "\n", "\n", "\n", "\n", "43\n", "\n", " \n", "tomato\n", "node score:1.0000\n", "\n", "\n", "\n", "47->43\n", "\n", "\n", "\n", "\n", "\n" ], "text/plain": [ "" ] }, "metadata": {}, "output_type": "display_data" }, { "data": { "image/svg+xml": [ "\n", "\n", "\n", "\n", "\n", "\n", "%3\n", "\n", "\n", "\n", "64\n", "\n", " \n", "mix\n", "node score: 0.0000\n", "\n", "\n", "\n", "59\n", "\n", " \n", "beat\n", "node score: 0.7344\n", "\n", "\n", "\n", "64->59\n", "\n", "\n", "\n", "\n", "\n", "63\n", "\n", " \n", "bake\n", "node score: 0.4789\n", "\n", "\n", "\n", "64->63\n", "\n", "\n", "\n", "\n", "\n", "54\n", "\n", " \n", "egg\n", "node score:1.0000\n", "\n", "\n", "\n", "59->54\n", "\n", "\n", "\n", "\n", "\n", "62\n", "\n", " \n", "mix\n", "node score: 0.2000\n", "\n", "\n", "\n", "63->62\n", "\n", "\n", "\n", "\n", "\n", "61\n", "\n", " \n", "cook\n", "node score: 0.7853\n", "\n", "\n", "\n", "62->61\n", "\n", "\n", "\n", "\n", "\n", "55\n", "\n", " \n", "cheese\n", "node score:1.0000\n", "\n", "\n", "\n", "62->55\n", "\n", "\n", "\n", "\n", "\n", "60\n", "\n", " \n", "mix\n", "node score: 0.4286\n", "\n", "\n", "\n", "61->60\n", "\n", "\n", "\n", "\n", "\n", "52\n", "\n", " \n", "onion\n", "node score:1.0000\n", "\n", "\n", "\n", "60->52\n", "\n", "\n", "\n", "\n", "\n", "58\n", "\n", " \n", "heat\n", "node score: 1.0000\n", "\n", "\n", "\n", "60->58\n", "\n", "\n", "\n", "\n", "\n", "50\n", "\n", " \n", "bacon\n", "node score:1.0000\n", "\n", "\n", "\n", "60->50\n", "\n", "\n", "\n", "\n", "\n", "57\n", "\n", " \n", "mix\n", "node score: 1.0000\n", "\n", "\n", "\n", "58->57\n", "\n", "\n", "\n", "\n", "\n", "56\n", "\n", " \n", "olive oil\n", "node score:1.0000\n", "\n", "\n", "\n", "57->56\n", "\n", "\n", "\n", "\n", "\n", "51\n", "\n", " \n", "tomato\n", "node score:1.0000\n", "\n", "\n", "\n", "57->51\n", "\n", "\n", "\n", "\n", "\n", "53\n", "\n", " \n", "garlic clove\n", "node score:1.0000\n", "\n", "\n", "\n", "57->53\n", "\n", "\n", "\n", "\n", "\n" ], "text/plain": [ "" ] }, "metadata": {}, "output_type": "display_data" }, { "data": { "image/svg+xml": [ "\n", "\n", "\n", "\n", "\n", "\n", "%3\n", "\n", "\n", "\n", "74\n", "\n", " \n", "cook\n", "node score: 0.7125\n", "\n", "\n", "\n", "73\n", "\n", " \n", "mix\n", "node score: 0.2500\n", "\n", "\n", "\n", "74->73\n", "\n", "\n", "\n", "\n", "\n", "66\n", "\n", " \n", "bacon\n", "node score:1.0000\n", "\n", "\n", "\n", "73->66\n", "\n", "\n", "\n", "\n", "\n", "72\n", "\n", " \n", "heat\n", "node score: 0.8920\n", "\n", "\n", "\n", "73->72\n", "\n", "\n", "\n", "\n", "\n", "71\n", "\n", " \n", "mix\n", "node score: 1.0000\n", "\n", "\n", "\n", "72->71\n", "\n", "\n", "\n", "\n", "\n", "68\n", "\n", " \n", "onion\n", "node score:1.0000\n", "\n", "\n", "\n", "71->68\n", "\n", "\n", "\n", "\n", "\n", "69\n", "\n", " \n", "salt\n", "node score:1.0000\n", "\n", "\n", "\n", "71->69\n", "\n", "\n", "\n", "\n", "\n", "70\n", "\n", " \n", "olive oil\n", "node score:1.0000\n", "\n", "\n", "\n", "71->70\n", "\n", "\n", "\n", "\n", "\n", "67\n", "\n", " \n", "tomato\n", "node score:1.0000\n", "\n", "\n", "\n", "71->67\n", "\n", "\n", "\n", "\n", "\n" ], "text/plain": [ "" ] }, "metadata": {}, "output_type": "display_data" }, { "data": { "image/svg+xml": [ "\n", "\n", "\n", "\n", "\n", "\n", "%3\n", "\n", "\n", "\n", "82\n", "\n", " \n", "mix\n", "node score: 0.0000\n", "\n", "\n", "\n", "81\n", "\n", " \n", "heat\n", "node score: 1.0000\n", "\n", "\n", "\n", "82->81\n", "\n", "\n", "\n", "\n", "\n", "80\n", "\n", " \n", "cook\n", "node score: 1.0000\n", "\n", "\n", "\n", "82->80\n", "\n", "\n", "\n", "\n", "\n", "77\n", "\n", " \n", "tomato\n", "node score:1.0000\n", "\n", "\n", "\n", "81->77\n", "\n", "\n", "\n", "\n", "\n", "79\n", "\n", " \n", "mix\n", "node score: 1.0000\n", "\n", "\n", "\n", "80->79\n", "\n", "\n", "\n", "\n", "\n", "76\n", "\n", " \n", "bacon\n", "node score:1.0000\n", "\n", "\n", "\n", "79->76\n", "\n", "\n", "\n", "\n", "\n", "78\n", "\n", " \n", "onion\n", "node score:1.0000\n", "\n", "\n", "\n", "79->78\n", "\n", "\n", "\n", "\n", "\n" ], "text/plain": [ "" ] }, "metadata": {}, "output_type": "display_data" }, { "data": { "image/svg+xml": [ "\n", "\n", "\n", "\n", "\n", "\n", "%3\n", "\n", "\n", "\n", "97\n", "\n", " \n", "mix\n", "node score: 0.0000\n", "\n", "\n", "\n", "96\n", "\n", " \n", "heat\n", "node score: 0.9279\n", "\n", "\n", "\n", "97->96\n", "\n", "\n", "\n", "\n", "\n", "94\n", "\n", " \n", "cool\n", "node score: 0.6073\n", "\n", "\n", "\n", "97->94\n", "\n", "\n", "\n", "\n", "\n", "95\n", "\n", " \n", "mix\n", "node score: 0.5000\n", "\n", "\n", "\n", "96->95\n", "\n", "\n", "\n", "\n", "\n", "92\n", "\n", " \n", "cook\n", "node score: 0.7929\n", "\n", "\n", "\n", "95->92\n", "\n", "\n", "\n", "\n", "\n", "85\n", "\n", " \n", "tomato\n", "node score:1.0000\n", "\n", "\n", "\n", "95->85\n", "\n", "\n", "\n", "\n", "\n", "91\n", "\n", " \n", "mix\n", "node score: 0.8333\n", "\n", "\n", "\n", "92->91\n", "\n", "\n", "\n", "\n", "\n", "84\n", "\n", " \n", "bacon\n", "node score:1.0000\n", "\n", "\n", "\n", "91->84\n", "\n", "\n", "\n", "\n", "\n", "86\n", "\n", " \n", "onion\n", "node score:1.0000\n", "\n", "\n", "\n", "91->86\n", "\n", "\n", "\n", "\n", "\n", "88\n", "\n", " \n", "olive oil\n", "node score:1.0000\n", "\n", "\n", "\n", "91->88\n", "\n", "\n", "\n", "\n", "\n", "89\n", "\n", " \n", "garlic clove\n", "node score:1.0000\n", "\n", "\n", "\n", "91->89\n", "\n", "\n", "\n", "\n", "\n", "93\n", "\n", " \n", "mix\n", "node score: 1.0000\n", "\n", "\n", "\n", "94->93\n", "\n", "\n", "\n", "\n", "\n", "87\n", "\n", " \n", "salt\n", "node score:1.0000\n", "\n", "\n", "\n", "93->87\n", "\n", "\n", "\n", "\n", "\n", "90\n", "\n", " \n", "egg\n", "node score:1.0000\n", "\n", "\n", "\n", "93->90\n", "\n", "\n", "\n", "\n", "\n" ], "text/plain": [ "" ] }, "metadata": {}, "output_type": "display_data" }, { "data": { "image/svg+xml": [ "\n", "\n", "\n", "\n", "\n", "\n", "%3\n", "\n", "\n", "\n", "108\n", "\n", " \n", "mix\n", "node score: 0.0000\n", "\n", "\n", "\n", "107\n", "\n", " \n", "cook\n", "node score: 0.9182\n", "\n", "\n", "\n", "108->107\n", "\n", "\n", "\n", "\n", "\n", "103\n", "\n", " \n", "cool\n", "node score: 0.5490\n", "\n", "\n", "\n", "108->103\n", "\n", "\n", "\n", "\n", "\n", "106\n", "\n", " \n", "mix\n", "node score: 0.0000\n", "\n", "\n", "\n", "107->106\n", "\n", "\n", "\n", "\n", "\n", "99\n", "\n", " \n", "bacon\n", "node score:1.0000\n", "\n", "\n", "\n", "106->99\n", "\n", "\n", "\n", "\n", "\n", "105\n", "\n", " \n", "simmer\n", "node score: 0.6840\n", "\n", "\n", "\n", "106->105\n", "\n", "\n", "\n", "\n", "\n", "104\n", "\n", " \n", "mix\n", "node score: 1.0000\n", "\n", "\n", "\n", "105->104\n", "\n", "\n", "\n", "\n", "\n", "100\n", "\n", " \n", "tomato\n", "node score:1.0000\n", "\n", "\n", "\n", "104->100\n", "\n", "\n", "\n", "\n", "\n", "101\n", "\n", " \n", "onion\n", "node score:1.0000\n", "\n", "\n", "\n", "104->101\n", "\n", "\n", "\n", "\n", "\n", "102\n", "\n", " \n", "salt\n", "node score:1.0000\n", "\n", "\n", "\n", "103->102\n", "\n", "\n", "\n", "\n", "\n" ], "text/plain": [ "" ] }, "metadata": {}, "output_type": "display_data" } ], "source": [ "p.plot_population(collect_scores=False)" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [] } ], "metadata": { "file_extension": ".py", "kernelspec": { "display_name": "Python 3", "language": "python", "name": "python3" }, "language_info": { "codemirror_mode": { "name": "ipython", "version": 3 }, "file_extension": ".py", "mimetype": "text/x-python", "name": "python", "nbconvert_exporter": "python", "pygments_lexer": "ipython3", "version": "3.7.5" }, "mimetype": "text/x-python", "name": "python", "npconvert_exporter": "python", "pygments_lexer": "ipython3", "version": 3 }, "nbformat": 4, "nbformat_minor": 4 }