Loading deepchem/models/tensorgraph/models/graph_models.py +1 −1 Original line number Diff line number Diff line Loading @@ -1112,7 +1112,7 @@ class MPNNTensorGraph(TensorGraph): results = [] for feed_dict in generator: # Extract number of unique samples in the batch from w_b n_valid_samples = len(np.nonzero(feed_dict[self.weights][:, 0])[0]) n_valid_samples = len(np.nonzero(np.sum(feed_dict[self.weights], 1))[0]) feed_dict = { self.layers[k.name].out_tensor: v for k, v in six.iteritems(feed_dict) Loading deepchem/molnet/actor_critic.py 0 → 100644 +126 −0 Original line number Diff line number Diff line #!/usr/bin/env python2 # -*- coding: utf-8 -*- """ Created on Thu Nov 2 16:27:33 2017 @author: zqwu """ from __future__ import division import numpy as np import tensorflow as tf import os from sklearn.externals import joblib import sklearn.pipeline import sklearn.preprocessing from sklearn.kernel_approximation import RBFSampler class Actor(object): def __init__(self, len_states=10, n_actions=7, batch_size=32, lr=0.001): self.len_states = len_states self.n_actions = n_actions # Number of actions self.batch_size = batch_size self.lr = lr # learning rate self.initialize_featurizer() self.g = tf.Graph() with self.g.as_default(): self._build_model() self._add_train_cost() self.sess = tf.Session(graph=self.g) self.sess.run(tf.global_variables_initializer()) def save(self, model_path, global_step=1): """ Save model """ with self.g.as_default(): saver = tf.train.Saver() saver.save(self.sess, os.path.join(model_path, "model"), global_step) joblib.dump(self.scaler, os.path.join(model_path, "scaler.joblib")) joblib.dump(self.featurizer, os.path.join(model_path, "featurizer.joblib")) def restore(self, model_path, global_step=1): """ Restore model """ with self.g.as_default(): saver = tf.train.Saver() saver.restore(self.sess, os.path.join(model_path, "model-") + str(global_step)) self.scaler = joblib.load(os.path.join(model_path, "scaler.joblib")) self.featurizer = joblib.load(os.path.join(model_path, "featurizer.joblib")) def initialize_featurizer(self): observation_examples = np.array([np.random.uniform(-np.pi, np.pi, (self.len_states,)) for x in range(10000)]) self.scaler = sklearn.preprocessing.StandardScaler() self.scaler.fit(observation_examples) # Normalization self.featurizer = sklearn.pipeline.FeatureUnion([ ("rbf1", RBFSampler(gamma=5.0, n_components=100)), ("rbf2", RBFSampler(gamma=2.0, n_components=100)), ("rbf3", RBFSampler(gamma=1.0, n_components=100)), ("rbf4", RBFSampler(gamma=0.5, n_components=100))]) self.featurizer.fit(self.scaler.transform(observation_examples)) # RBF featurizer def process_states(self, s): # This function featurize states(batch_size * 2) into inputs(batch_size * 400) assert len(s.shape) == 2 assert s.shape[1] == self.len_states scaled = self.scaler.transform(s) featurized = self.featurizer.transform(scaled) return featurized def _build_model(self): # Inputs: batch_size * 400 self.state = tf.placeholder(tf.float32, shape=(None, 400)) # First hidden layer: batch_size * 10 W1 = tf.Variable(np.random.normal(0, 0.02, (400, 10)), dtype=tf.float32) b1 = tf.Variable(np.zeros((10,))+0.01, dtype=tf.float32) hidden1 = tf.nn.tanh(tf.matmul(self.state, W1) + b1) # State values: batch_size * 1 W2 = tf.Variable(np.random.normal(0, 0.02, (10, 1)), dtype=tf.float32) b2 = tf.Variable(np.zeros((1,))+0.01, dtype=tf.float32) self.V = tf.matmul(hidden1, W2) + b2 # Action values: batch_size * n_actions W3 = tf.Variable(np.random.normal(0, 0.02, (10, self.n_actions)), dtype=tf.float32) b3 = tf.Variable(np.zeros((self.n_actions,))+0.01, dtype=tf.float32) self.A = tf.matmul(hidden1, W3) + b3 W4 = tf.Variable(np.random.normal(0, 0.02, (10, self.n_actions)), dtype=tf.float32) b4 = tf.Variable(np.zeros((self.n_actions,))+0.01, dtype=tf.float32) self.sigma = tf.matmul(hidden1, W4) + b4 # State-action values: batch_size * n_actions self.mu = self.V + (self.A - tf.reduce_mean(self.A, axis=1, keep_dims=True)) self.normal_dist = tf.contrib.distributions.Normal(tf.reshape(self.mu, (-1,)), tf.reshape(self.sigma, (-1,))) self.a = tf.reshape(self.normal_dist.sample(), (-1, self.n_actions)) def _add_train_cost(self): # Target Q values self.action = tf.placeholder(tf.float32, shape=(None, self.n_actions)) self.target = tf.placeholder(tf.float32, shape=(None,)) target = tf.stack([self.target]*self.n_actions, axis=1) # Minimize L2 loss self.loss = -self.normal_dist.log_prob(tf.reshape(self.action, (-1,))) * tf.reshape(target, (-1,)) self.loss -= 0.01 * self.normal_dist.entropy() self.optimizer = tf.train.AdamOptimizer(learning_rate=self.lr) self.train_op = self.optimizer.minimize(tf.reduce_sum(self.loss)) def update_Q(self, states, action, target): # Update model variables s = np.reshape(np.array(states, dtype=float), (-1, self.len_states)) feed_dict = { self.state: self.process_states(s), self.action: action, self.target: target } self.sess.run(self.train_op, feed_dict=feed_dict) def ChooseAction(self, states, mode='epsilon', epsilon=0.9): # Using epsilon-greedy to choose actions s = np.reshape(np.array(states, dtype=float), (-1, self.len_states)) feed_dict = {self.state: self.process_states(s)} a = self.sess.run(self.a, feed_dict=feed_dict) return a No newline at end of file deepchem/molnet/check_availability.py +8 −0 Original line number Diff line number Diff line Loading @@ -10,6 +10,7 @@ CheckFeaturizer = { ('bace_c', 'dag'): ['GraphConv', 75], ('bace_c', 'weave'): ['Weave', 75], ('bace_c', 'textcnn'): ['Raw', None], ('bace_c', 'mpnn'): ['Weave', [75, 14]], ('bbbp', 'logreg'): ['ECFP', 1024], ('bbbp', 'tf'): ['ECFP', 1024], ('bbbp', 'tf_robust'): ['ECFP', 1024], Loading @@ -21,6 +22,7 @@ CheckFeaturizer = { ('bbbp', 'dag'): ['GraphConv', 75], ('bbbp', 'weave'): ['Weave', 75], ('bbbp', 'textcnn'): ['Raw', None], ('bbbp', 'mpnn'): ['Weave', [75, 14]], ('clintox', 'logreg'): ['ECFP', 1024], ('clintox', 'tf'): ['ECFP', 1024], ('clintox', 'tf_robust'): ['ECFP', 1024], Loading @@ -32,6 +34,7 @@ CheckFeaturizer = { ('clintox', 'dag'): ['GraphConv', 75], ('clintox', 'weave'): ['Weave', 75], ('clintox', 'textcnn'): ['Raw', None], ('clintox', 'mpnn'): ['Weave', [75, 14]], ('hiv', 'logreg'): ['ECFP', 1024], ('hiv', 'tf'): ['ECFP', 1024], ('hiv', 'tf_robust'): ['ECFP', 1024], Loading @@ -43,6 +46,7 @@ CheckFeaturizer = { ('hiv', 'dag'): ['GraphConv', 75], ('hiv', 'weave'): ['Weave', 75], ('hiv', 'textcnn'): ['Raw', None], ('hiv', 'mpnn'): ['Weave', [75, 14]], ('muv', 'logreg'): ['ECFP', 1024], ('muv', 'tf'): ['ECFP', 1024], ('muv', 'tf_robust'): ['ECFP', 1024], Loading @@ -56,6 +60,7 @@ CheckFeaturizer = { ('muv', 'res'): ['GraphConv', 75], ('muv', 'weave'): ['Weave', 75], ('muv', 'textcnn'): ['Raw', None], ('muv', 'mpnn'): ['Weave', [75, 14]], ('pcba', 'logreg'): ['ECFP', 1024], ('pcba', 'tf'): ['ECFP', 1024], ('pcba', 'tf_robust'): ['ECFP', 1024], Loading Loading @@ -92,6 +97,7 @@ CheckFeaturizer = { ('sider', 'attn'): ['GraphConv', 75], ('sider', 'res'): ['GraphConv', 75], ('sider', 'textcnn'): ['Raw', None], ('sider', 'mpnn'): ['Weave', [75, 14]], ('tox21', 'logreg'): ['ECFP', 1024], ('tox21', 'tf'): ['ECFP', 1024], ('tox21', 'tf_robust'): ['ECFP', 1024], Loading @@ -106,6 +112,7 @@ CheckFeaturizer = { ('tox21', 'attn'): ['GraphConv', 75], ('tox21', 'res'): ['GraphConv', 75], ('tox21', 'textcnn'): ['Raw', None], ('tox21', 'mpnn'): ['Weave', [75, 14]], ('toxcast', 'logreg'): ['ECFP', 1024], ('toxcast', 'tf'): ['ECFP', 1024], ('toxcast', 'tf_robust'): ['ECFP', 1024], Loading @@ -116,6 +123,7 @@ CheckFeaturizer = { ('toxcast', 'graphconv'): ['GraphConv', 75], ('toxcast', 'weave'): ['Weave', 75], ('toxcast', 'textcnn'): ['Raw', None], ('toxcast', 'mpnn'): ['Weave', [75, 14]], ('bace_r', 'tf_regression'): ['ECFP', 1024], ('bace_r', 'rf_regression'): ['ECFP', 1024], ('bace_r', 'krr'): ['ECFP', 1024], Loading deepchem/molnet/preset_hyper_parameters.py +8 −8 Original line number Diff line number Diff line Loading @@ -83,6 +83,14 @@ hps['textcnn'] = { 'num_filters': [100, 200, 200, 200, 200, 100, 100, 100, 100, 100, 160, 160], 'seed': 123 } hps['mpnn'] = { 'batch_size': 16, 'nb_epoch': 50, 'learning_rate': 0.0005, 'T': 2, 'M': 5, 'seed': 123 } hps['rf'] = {'n_estimators': 500} hps['kernelsvm'] = {'C': 1.0, 'gamma': 0.05} hps['xgb'] = { Loading Loading @@ -177,14 +185,6 @@ hps['ani'] = { 'layer_structures': [20, 10, 10], 'seed': 123 } hps['mpnn'] = { 'batch_size': 64, 'nb_epoch': 50, 'learning_rate': 0.001, 'T': 2, 'M': 5, 'seed': 123 } hps['xgb_regression'] = { 'max_depth': 5, Loading deepchem/molnet/qm7_tg_GraphConv_data/weights/checkpoint 0 → 100644 +8 −0 Original line number Diff line number Diff line model_checkpoint_path: "model-699" all_model_checkpoint_paths: "model-99" all_model_checkpoint_paths: "model-199" all_model_checkpoint_paths: "model-299" all_model_checkpoint_paths: "model-399" all_model_checkpoint_paths: "model-499" all_model_checkpoint_paths: "model-599" all_model_checkpoint_paths: "model-699" Loading
deepchem/models/tensorgraph/models/graph_models.py +1 −1 Original line number Diff line number Diff line Loading @@ -1112,7 +1112,7 @@ class MPNNTensorGraph(TensorGraph): results = [] for feed_dict in generator: # Extract number of unique samples in the batch from w_b n_valid_samples = len(np.nonzero(feed_dict[self.weights][:, 0])[0]) n_valid_samples = len(np.nonzero(np.sum(feed_dict[self.weights], 1))[0]) feed_dict = { self.layers[k.name].out_tensor: v for k, v in six.iteritems(feed_dict) Loading
deepchem/molnet/actor_critic.py 0 → 100644 +126 −0 Original line number Diff line number Diff line #!/usr/bin/env python2 # -*- coding: utf-8 -*- """ Created on Thu Nov 2 16:27:33 2017 @author: zqwu """ from __future__ import division import numpy as np import tensorflow as tf import os from sklearn.externals import joblib import sklearn.pipeline import sklearn.preprocessing from sklearn.kernel_approximation import RBFSampler class Actor(object): def __init__(self, len_states=10, n_actions=7, batch_size=32, lr=0.001): self.len_states = len_states self.n_actions = n_actions # Number of actions self.batch_size = batch_size self.lr = lr # learning rate self.initialize_featurizer() self.g = tf.Graph() with self.g.as_default(): self._build_model() self._add_train_cost() self.sess = tf.Session(graph=self.g) self.sess.run(tf.global_variables_initializer()) def save(self, model_path, global_step=1): """ Save model """ with self.g.as_default(): saver = tf.train.Saver() saver.save(self.sess, os.path.join(model_path, "model"), global_step) joblib.dump(self.scaler, os.path.join(model_path, "scaler.joblib")) joblib.dump(self.featurizer, os.path.join(model_path, "featurizer.joblib")) def restore(self, model_path, global_step=1): """ Restore model """ with self.g.as_default(): saver = tf.train.Saver() saver.restore(self.sess, os.path.join(model_path, "model-") + str(global_step)) self.scaler = joblib.load(os.path.join(model_path, "scaler.joblib")) self.featurizer = joblib.load(os.path.join(model_path, "featurizer.joblib")) def initialize_featurizer(self): observation_examples = np.array([np.random.uniform(-np.pi, np.pi, (self.len_states,)) for x in range(10000)]) self.scaler = sklearn.preprocessing.StandardScaler() self.scaler.fit(observation_examples) # Normalization self.featurizer = sklearn.pipeline.FeatureUnion([ ("rbf1", RBFSampler(gamma=5.0, n_components=100)), ("rbf2", RBFSampler(gamma=2.0, n_components=100)), ("rbf3", RBFSampler(gamma=1.0, n_components=100)), ("rbf4", RBFSampler(gamma=0.5, n_components=100))]) self.featurizer.fit(self.scaler.transform(observation_examples)) # RBF featurizer def process_states(self, s): # This function featurize states(batch_size * 2) into inputs(batch_size * 400) assert len(s.shape) == 2 assert s.shape[1] == self.len_states scaled = self.scaler.transform(s) featurized = self.featurizer.transform(scaled) return featurized def _build_model(self): # Inputs: batch_size * 400 self.state = tf.placeholder(tf.float32, shape=(None, 400)) # First hidden layer: batch_size * 10 W1 = tf.Variable(np.random.normal(0, 0.02, (400, 10)), dtype=tf.float32) b1 = tf.Variable(np.zeros((10,))+0.01, dtype=tf.float32) hidden1 = tf.nn.tanh(tf.matmul(self.state, W1) + b1) # State values: batch_size * 1 W2 = tf.Variable(np.random.normal(0, 0.02, (10, 1)), dtype=tf.float32) b2 = tf.Variable(np.zeros((1,))+0.01, dtype=tf.float32) self.V = tf.matmul(hidden1, W2) + b2 # Action values: batch_size * n_actions W3 = tf.Variable(np.random.normal(0, 0.02, (10, self.n_actions)), dtype=tf.float32) b3 = tf.Variable(np.zeros((self.n_actions,))+0.01, dtype=tf.float32) self.A = tf.matmul(hidden1, W3) + b3 W4 = tf.Variable(np.random.normal(0, 0.02, (10, self.n_actions)), dtype=tf.float32) b4 = tf.Variable(np.zeros((self.n_actions,))+0.01, dtype=tf.float32) self.sigma = tf.matmul(hidden1, W4) + b4 # State-action values: batch_size * n_actions self.mu = self.V + (self.A - tf.reduce_mean(self.A, axis=1, keep_dims=True)) self.normal_dist = tf.contrib.distributions.Normal(tf.reshape(self.mu, (-1,)), tf.reshape(self.sigma, (-1,))) self.a = tf.reshape(self.normal_dist.sample(), (-1, self.n_actions)) def _add_train_cost(self): # Target Q values self.action = tf.placeholder(tf.float32, shape=(None, self.n_actions)) self.target = tf.placeholder(tf.float32, shape=(None,)) target = tf.stack([self.target]*self.n_actions, axis=1) # Minimize L2 loss self.loss = -self.normal_dist.log_prob(tf.reshape(self.action, (-1,))) * tf.reshape(target, (-1,)) self.loss -= 0.01 * self.normal_dist.entropy() self.optimizer = tf.train.AdamOptimizer(learning_rate=self.lr) self.train_op = self.optimizer.minimize(tf.reduce_sum(self.loss)) def update_Q(self, states, action, target): # Update model variables s = np.reshape(np.array(states, dtype=float), (-1, self.len_states)) feed_dict = { self.state: self.process_states(s), self.action: action, self.target: target } self.sess.run(self.train_op, feed_dict=feed_dict) def ChooseAction(self, states, mode='epsilon', epsilon=0.9): # Using epsilon-greedy to choose actions s = np.reshape(np.array(states, dtype=float), (-1, self.len_states)) feed_dict = {self.state: self.process_states(s)} a = self.sess.run(self.a, feed_dict=feed_dict) return a No newline at end of file
deepchem/molnet/check_availability.py +8 −0 Original line number Diff line number Diff line Loading @@ -10,6 +10,7 @@ CheckFeaturizer = { ('bace_c', 'dag'): ['GraphConv', 75], ('bace_c', 'weave'): ['Weave', 75], ('bace_c', 'textcnn'): ['Raw', None], ('bace_c', 'mpnn'): ['Weave', [75, 14]], ('bbbp', 'logreg'): ['ECFP', 1024], ('bbbp', 'tf'): ['ECFP', 1024], ('bbbp', 'tf_robust'): ['ECFP', 1024], Loading @@ -21,6 +22,7 @@ CheckFeaturizer = { ('bbbp', 'dag'): ['GraphConv', 75], ('bbbp', 'weave'): ['Weave', 75], ('bbbp', 'textcnn'): ['Raw', None], ('bbbp', 'mpnn'): ['Weave', [75, 14]], ('clintox', 'logreg'): ['ECFP', 1024], ('clintox', 'tf'): ['ECFP', 1024], ('clintox', 'tf_robust'): ['ECFP', 1024], Loading @@ -32,6 +34,7 @@ CheckFeaturizer = { ('clintox', 'dag'): ['GraphConv', 75], ('clintox', 'weave'): ['Weave', 75], ('clintox', 'textcnn'): ['Raw', None], ('clintox', 'mpnn'): ['Weave', [75, 14]], ('hiv', 'logreg'): ['ECFP', 1024], ('hiv', 'tf'): ['ECFP', 1024], ('hiv', 'tf_robust'): ['ECFP', 1024], Loading @@ -43,6 +46,7 @@ CheckFeaturizer = { ('hiv', 'dag'): ['GraphConv', 75], ('hiv', 'weave'): ['Weave', 75], ('hiv', 'textcnn'): ['Raw', None], ('hiv', 'mpnn'): ['Weave', [75, 14]], ('muv', 'logreg'): ['ECFP', 1024], ('muv', 'tf'): ['ECFP', 1024], ('muv', 'tf_robust'): ['ECFP', 1024], Loading @@ -56,6 +60,7 @@ CheckFeaturizer = { ('muv', 'res'): ['GraphConv', 75], ('muv', 'weave'): ['Weave', 75], ('muv', 'textcnn'): ['Raw', None], ('muv', 'mpnn'): ['Weave', [75, 14]], ('pcba', 'logreg'): ['ECFP', 1024], ('pcba', 'tf'): ['ECFP', 1024], ('pcba', 'tf_robust'): ['ECFP', 1024], Loading Loading @@ -92,6 +97,7 @@ CheckFeaturizer = { ('sider', 'attn'): ['GraphConv', 75], ('sider', 'res'): ['GraphConv', 75], ('sider', 'textcnn'): ['Raw', None], ('sider', 'mpnn'): ['Weave', [75, 14]], ('tox21', 'logreg'): ['ECFP', 1024], ('tox21', 'tf'): ['ECFP', 1024], ('tox21', 'tf_robust'): ['ECFP', 1024], Loading @@ -106,6 +112,7 @@ CheckFeaturizer = { ('tox21', 'attn'): ['GraphConv', 75], ('tox21', 'res'): ['GraphConv', 75], ('tox21', 'textcnn'): ['Raw', None], ('tox21', 'mpnn'): ['Weave', [75, 14]], ('toxcast', 'logreg'): ['ECFP', 1024], ('toxcast', 'tf'): ['ECFP', 1024], ('toxcast', 'tf_robust'): ['ECFP', 1024], Loading @@ -116,6 +123,7 @@ CheckFeaturizer = { ('toxcast', 'graphconv'): ['GraphConv', 75], ('toxcast', 'weave'): ['Weave', 75], ('toxcast', 'textcnn'): ['Raw', None], ('toxcast', 'mpnn'): ['Weave', [75, 14]], ('bace_r', 'tf_regression'): ['ECFP', 1024], ('bace_r', 'rf_regression'): ['ECFP', 1024], ('bace_r', 'krr'): ['ECFP', 1024], Loading
deepchem/molnet/preset_hyper_parameters.py +8 −8 Original line number Diff line number Diff line Loading @@ -83,6 +83,14 @@ hps['textcnn'] = { 'num_filters': [100, 200, 200, 200, 200, 100, 100, 100, 100, 100, 160, 160], 'seed': 123 } hps['mpnn'] = { 'batch_size': 16, 'nb_epoch': 50, 'learning_rate': 0.0005, 'T': 2, 'M': 5, 'seed': 123 } hps['rf'] = {'n_estimators': 500} hps['kernelsvm'] = {'C': 1.0, 'gamma': 0.05} hps['xgb'] = { Loading Loading @@ -177,14 +185,6 @@ hps['ani'] = { 'layer_structures': [20, 10, 10], 'seed': 123 } hps['mpnn'] = { 'batch_size': 64, 'nb_epoch': 50, 'learning_rate': 0.001, 'T': 2, 'M': 5, 'seed': 123 } hps['xgb_regression'] = { 'max_depth': 5, Loading
deepchem/molnet/qm7_tg_GraphConv_data/weights/checkpoint 0 → 100644 +8 −0 Original line number Diff line number Diff line model_checkpoint_path: "model-699" all_model_checkpoint_paths: "model-99" all_model_checkpoint_paths: "model-199" all_model_checkpoint_paths: "model-299" all_model_checkpoint_paths: "model-399" all_model_checkpoint_paths: "model-499" all_model_checkpoint_paths: "model-599" all_model_checkpoint_paths: "model-699"
