Loading deepchem/metrics/__init__.py +15 −4 Original line number Diff line number Diff line Loading @@ -123,7 +123,7 @@ class Metric(object): assert mode in ["classification", "regression"] self.mode = mode def compute_metric(self, y_true, y_pred, w): def compute_metric(self, y_true, y_pred, w, n_classes=2): """Compute a performance metric for each task. Args: Loading @@ -140,7 +140,10 @@ class Metric(object): computed_metrics = [] for task in xrange(num_tasks): y_task = y_true[:, task] if self.mode == "regression": y_pred_task = y_pred[:, task] else: y_pred_task = y_pred[:, task*n_classes:(task+1)*n_classes] w_task = w[:, task] try: Loading Loading @@ -172,14 +175,22 @@ class Metric(object): Raises: NotImplementedError: If metric_str is not in METRICS. """ y_true = y_true[w != 0] y_pred = y_pred[w != 0] y_true = np.squeeze(y_true[w != 0]) y_pred = np.squeeze(y_pred[w != 0]) # If there are no nonzero examples, metric is ill-defined. if not len(y_true): return np.nan print("y_pred") print(y_pred) if self.mode == "classification": if self.name == "roc_auc_score": y_true = to_one_hot(y_true).astype(int) y_pred = y_pred[:, np.newaxis] else: y_true = y_true.astype(int) y_pred = from_one_hot(y_pred) #y_pred = from_one_hot(y_pred[:, np.newaxis]) print("y_pred") print(y_pred) if self.threshold is not None: y_pred = np.greater(y_pred, threshold) try: Loading deepchem/models/__init__.py +4 −0 Original line number Diff line number Diff line Loading @@ -138,6 +138,10 @@ class Model(object): y_preds = [] batch_size = self.model_params["batch_size"] for (X_batch, y_batch, w_batch, ids_batch) in dataset.iterbatches(batch_size): print("y_batch.shape") print(y_batch.shape) print("self.predict_on_batch(X_batch).shape") print(self.predict_on_batch(X_batch).shape) y_pred_batch = np.reshape(self.predict_on_batch(X_batch), y_batch.shape) y_pred_batch = undo_transforms(y_pred_batch, transformers) y_preds.append(y_pred_batch) Loading deepchem/models/keras_models/fcnet.py +17 −0 Original line number Diff line number Diff line Loading @@ -133,6 +133,23 @@ class MultiTaskDNN(KerasModel): y_pred = np.squeeze(y_pred) return y_pred def predict_proba_on_batch(self, X, n_classes=2): """ Makes predictions on given batch of new data. """ data = self.get_data_dict(X) y_pred_dict = self.raw_model.predict_on_batch(data) nb_samples = np.shape(X)[0] nb_tasks = len(self.tasks) y_pred = np.zeros((nb_samples, n_classes*nb_tasks)) for ind, task in enumerate(self.tasks): task_type = self.task_types[task] taskname = "task%d" % ind y_pred_task = np.squeeze(y_pred_dict[taskname]) y_pred[:, ind:ind+n_classes] = y_pred_task y_pred = np.squeeze(y_pred) return y_pred class SingleTaskDNN(MultiTaskDNN): """ Abstract base class for different ML models. Loading deepchem/models/tensorflow_models/__init__.py +38 −16 Original line number Diff line number Diff line Loading @@ -95,6 +95,7 @@ class TensorflowGraph(object): This function constructs the computational graph for the model. It relies subclassed methods (build/cost) to construct specific graphs. """ print("TensorflowGraph.__init__()") self.graph = tf.Graph() self.model_params = model_params self.logdir = logdir Loading Loading @@ -136,6 +137,7 @@ class TensorflowGraph(object): def setup(self): """Add ops common to training/eval to the graph.""" print("TensorflowGraph.setup()") with self.graph.as_default(): with tf.name_scope('core_model'): self.build() Loading @@ -160,13 +162,15 @@ class TensorflowGraph(object): return tf.name_scope(self._name_scopes[name]) def add_training_cost(self): print("TensorflowGraph.add_training_cost()") print("self.output") print(self.output) with self.graph.as_default(): self.require_attributes(['output', 'labels', 'weights']) epsilon = 1e-3 # small float to avoid dividing by zero model_params = self.model_params weighted_costs = [] # weighted costs for each example gradient_costs = [] # costs used for gradient calculation old_costs = [] # old-style cost with self._shared_name_scope('costs'): for task in xrange(self.num_tasks): Loading @@ -182,28 +186,23 @@ class TensorflowGraph(object): # non-zero weight examples in the batch. Also, instead of using # tf.reduce_mean (which can put ops on the CPU) we explicitly # calculate with div/sum so it stays on the GPU. print("model_params['batch_size']") print(model_params['batch_size']) print("weighted_cost") print(weighted_cost) gradient_cost = tf.div(tf.reduce_sum(weighted_cost), model_params["batch_size"]) gradient_costs.append(gradient_cost) with tf.name_scope('old_cost'): old_cost = tf.div( tf.reduce_sum(weighted_cost), tf.reduce_sum(self.weights[task]) + epsilon) old_costs.append(old_cost) # aggregated costs with self._shared_name_scope('aggregated'): with tf.name_scope('gradient'): loss = tf.add_n(gradient_costs) with tf.name_scope('old_cost'): old_loss = tf.add_n(old_costs) # weight decay if model_params["penalty"] != 0.0: penalty = model_ops.WeightDecay(model_params) loss += penalty old_loss += penalty # loss used for gradient calculation self.loss = loss Loading Loading @@ -255,9 +254,22 @@ class TensorflowGraph(object): for (X_b, y_b, w_b, ids_b) in dataset.iterbatches(batch_size): # Run training op and compute summaries. feed_dict = self.construct_feed_dict(X_b, y_b, w_b, ids_b) step, loss, _ = sess.run( [train_op.values()[0], self.loss, self.updates], #step, loss, _ = sess.run( # [train_op.values()[0], self.loss, self.updates], # feed_dict=feed_dict) output, step, loss, _ = sess.run( self.output + [train_op.values()[0], self.loss, self.updates], feed_dict=feed_dict) #print("loss") #print(loss) y_pred = np.squeeze(np.array(output)) #print("y_pred") #print(y_pred) y_b = y_b.flatten() #print("y_b") #print(y_b) #print(".5*np.sum((y_b - y_pred)**2)/len(y_b)") #print(.5*np.sum((y_b - y_pred)**2)/len(y_b)) # Save model checkpoints at end of epoch saver.save(sess, self._save_path, global_step=self.global_step) log('Ending epoch %d: loss %g' % (epoch, loss), self.verbosity) Loading Loading @@ -333,10 +345,13 @@ class TensorflowGraph(object): #labels = np.array(from_one_hot( # np.squeeze(np.concatenate(labels)), axis=-1)) labels = np.squeeze(np.concatenate(labels)) labels = np.array(labels)[:, 1] #labels = np.squeeze(np.concatenate(labels)) outputs = np.array(from_one_hot( np.squeeze(np.concatenate(output)), axis=-1)) #labels = np.array(labels)[:, 1] return np.copy(labels) #return np.copy(labels) return np.copy(outputs) def add_output_ops(self): """Replace logits with softmax outputs.""" Loading Loading @@ -658,7 +673,7 @@ class TensorflowRegressor(TensorflowGraph): A tensor with shape batch_size containing the weighted cost for each example. """ #return tf.mul(0.5 * tf.square(output - labels), weights) return tf.mul(0.5 * tf.square(output - labels), weights) def example_counts(self, y_true): """Get counts of examples in each class. Loading Loading @@ -734,6 +749,13 @@ class TensorflowModel(Model): """ return self.eval_model.predict_on_batch(X) def predict_proba_on_batch(self, X): """ Makes predictions on batch of data. """ return self.eval_model.predict_proba_on_batch(X) def save(self): """ No-op since tf models save themselves during fit() Loading deepchem/models/tensorflow_models/fcnet.py +117 −36 Original line number Diff line number Diff line Loading @@ -62,11 +62,22 @@ import numpy as np import tensorflow as tf from tensorflow.python.platform import logging from deepchem.metrics import from_one_hot from deepchem.models.tensorflow_models import TensorflowClassifier from deepchem.models.tensorflow_models import TensorflowRegressor from deepchem.models.tensorflow_models import model_ops from deepchem.metrics import to_one_hot def softmax(x): """Simple numpy softmax implementation """ tmp = np.max(x, axis = 1) x -= tmp.reshape((x.shape[0], 1)) x = np.exp(x) tmp = np.sum(x, axis = 1) x /= tmp.reshape((x.shape[0], 1)) return x class TensorflowMultiTaskClassifier(TensorflowClassifier): """Implements an icml model as configured in a model_config.proto.""" Loading Loading @@ -148,6 +159,92 @@ class TensorflowMultiTaskClassifier(TensorflowClassifier): orig_dict["valid"] = np.ones((self.model_params["batch_size"],), dtype=bool) return self._get_feed_dict(orig_dict) def predict_proba_on_batch(self, X): """Return model output for the provided input. Restore(checkpoint) must have previously been called on this object. Args: dataset: deepchem.datasets.dataset object. Returns: Tuple of three numpy arrays with shape num_examples x num_tasks (x ...): output: Model outputs. labels: True labels. weights: Example weights. Note that the output and labels arrays may be more than 2D, e.g. for classifier models that return class probabilities. Raises: AssertionError: If model is not in evaluation mode. ValueError: If output and labels are not both 3D or both 2D. """ if not self._restored_model: self.restore() with self.graph.as_default(): assert not model_ops.is_training() self.require_attributes(['output', 'labels', 'weights']) # run eval data through the model num_tasks = self.num_tasks outputs, labels, weights = [], [], [] start = time.time() with self._get_shared_session().as_default(): batch_count = -1.0 feed_dict = self.construct_feed_dict(X) batch_start = time.time() batch_count += 1 data = self._get_shared_session().run( self.output + self.labels + self.weights, feed_dict=feed_dict) batch_outputs = np.asarray(data[:num_tasks], dtype=float) batch_labels = np.asarray(data[num_tasks:num_tasks * 2], dtype=float) batch_weights = np.asarray(data[num_tasks * 2:num_tasks * 3], dtype=float) # reshape to batch_size x num_tasks x ... if batch_outputs.ndim == 3 and batch_labels.ndim == 3: batch_outputs = batch_outputs.transpose((1, 0, 2)) batch_labels = batch_labels.transpose((1, 0, 2)) elif batch_outputs.ndim == 2 and batch_labels.ndim == 2: batch_outputs = batch_outputs.transpose((1, 0)) batch_labels = batch_labels.transpose((1, 0)) else: raise ValueError( 'Unrecognized rank combination for output and labels: %s %s' % (batch_outputs.shape, batch_labels.shape)) batch_weights = batch_weights.transpose((1, 0)) valid = feed_dict[self.valid.name] # only take valid outputs if np.count_nonzero(~valid): batch_outputs = batch_outputs[valid] batch_labels = batch_labels[valid] batch_weights = batch_weights[valid] outputs.append(batch_outputs) labels.append(batch_labels) weights.append(batch_weights) print("predict_proba_on_batch()") print("batch_outputs") print(batch_outputs) print("batch_labels") print(batch_labels) logging.info('Eval batch took %g seconds', time.time() - start) #labels = np.array(from_one_hot( # np.squeeze(np.concatenate(labels)), axis=-1)) ##labels = np.squeeze(np.concatenate(labels)) #labels = np.array(labels)[:, 1] # We apply softmax to predictions to get class probabilities. #outputs = np.array(from_one_hot(softmax(np.squeeze(np.concatenate(outputs))))) outputs = softmax(np.squeeze(np.concatenate(outputs))) print("outputs") print(outputs) #return np.copy(labels) return np.copy(outputs) class TensorflowMultiTaskRegressor(TensorflowRegressor): """Implements an icml model as configured in a model_config.proto.""" Loading @@ -158,6 +255,7 @@ class TensorflowMultiTaskRegressor(TensorflowRegressor): mol_features: Molecule descriptor (e.g. fingerprint) tensor with shape batch_size x num_features. """ print("ENTERING TensorflowMultiTaskRegressor.build") assert len(self.model_params["data_shape"]) == 1 num_features = self.model_params["data_shape"][0] with self.graph.as_default(): Loading Loading @@ -197,14 +295,17 @@ class TensorflowMultiTaskRegressor(TensorflowRegressor): prev_layer = layer prev_layer_size = layer_sizes[i] self.output = [tf.squeeze(model_ops.FullyConnectedLayer( self.output = [] for task in range(self.num_tasks): self.output.append(tf.squeeze( model_ops.FullyConnectedLayer( tensor=prev_layer, size=layer_sizes[i], weight_init=tf.truncated_normal( shape=[prev_layer_size, 1], stddev=weight_init_stddevs[i]), bias_init=tf.constant(value=bias_init_consts[i], shape=[1])))] shape=[1])))) def construct_feed_dict(self, X_b, y_b=None, w_b=None, ids_b=None): """Construct a feed dictionary from minibatch data. Loading Loading @@ -259,52 +360,32 @@ class TensorflowMultiTaskRegressor(TensorflowRegressor): self.restore() with self.graph.as_default(): assert not model_ops.is_training() self.require_attributes(['output', 'labels', 'weights']) self.require_attributes(['output']) # run eval data through the model num_tasks = self.num_tasks output, labels, weights = [], [], [] start = time.time() outputs = [] with self._get_shared_session().as_default(): batch_count = -1.0 feed_dict = self.construct_feed_dict(X) batch_start = time.time() batch_count += 1 data = self._get_shared_session().run( self.output + self.labels + self.weights, feed_dict=feed_dict) batch_output = np.asarray(data[:num_tasks], dtype=float) batch_labels = np.asarray(data[num_tasks:num_tasks * 2], dtype=float) batch_weights = np.asarray(data[num_tasks * 2:num_tasks * 3], dtype=float) self.output, feed_dict=feed_dict) batch_outputs = np.asarray(data[:num_tasks], dtype=float) # reshape to batch_size x num_tasks x ... if batch_output.ndim == 3 and batch_labels.ndim == 3: batch_output = batch_output.transpose((1, 0, 2)) batch_labels = batch_labels.transpose((1, 0, 2)) elif batch_output.ndim == 2 and batch_labels.ndim == 2: batch_output = batch_output.transpose((1, 0)) batch_labels = batch_labels.transpose((1, 0)) if batch_outputs.ndim == 3: batch_outputs = batch_outputs.transpose((1, 0, 2)) elif batch_outputs.ndim == 2: batch_outputs = batch_outputs.transpose((1, 0)) else: raise ValueError( 'Unrecognized rank combination for output and labels: %s %s' % (batch_output.shape, batch_labels.shape)) batch_weights = batch_weights.transpose((1, 0)) 'Unrecognized rank combination for output: %s' % (batch_outputs.shape)) valid = feed_dict[self.valid.name] # only take valid outputs if np.count_nonzero(~valid): batch_output = batch_output[valid] batch_labels = batch_labels[valid] batch_weights = batch_weights[valid] output.append(batch_output) labels.append(batch_labels) weights.append(batch_weights) logging.info('Eval batch took %g seconds', time.time() - start) batch_outputs = batch_outputs[valid] outputs.append(batch_outputs) #labels = np.array(from_one_hot( # np.squeeze(np.concatenate(labels)), axis=-1)) labels = np.squeeze(np.concatenate(labels)) outputs = np.squeeze(np.concatenate(outputs)) return np.copy(labels) return np.copy(outputs) Loading
deepchem/metrics/__init__.py +15 −4 Original line number Diff line number Diff line Loading @@ -123,7 +123,7 @@ class Metric(object): assert mode in ["classification", "regression"] self.mode = mode def compute_metric(self, y_true, y_pred, w): def compute_metric(self, y_true, y_pred, w, n_classes=2): """Compute a performance metric for each task. Args: Loading @@ -140,7 +140,10 @@ class Metric(object): computed_metrics = [] for task in xrange(num_tasks): y_task = y_true[:, task] if self.mode == "regression": y_pred_task = y_pred[:, task] else: y_pred_task = y_pred[:, task*n_classes:(task+1)*n_classes] w_task = w[:, task] try: Loading Loading @@ -172,14 +175,22 @@ class Metric(object): Raises: NotImplementedError: If metric_str is not in METRICS. """ y_true = y_true[w != 0] y_pred = y_pred[w != 0] y_true = np.squeeze(y_true[w != 0]) y_pred = np.squeeze(y_pred[w != 0]) # If there are no nonzero examples, metric is ill-defined. if not len(y_true): return np.nan print("y_pred") print(y_pred) if self.mode == "classification": if self.name == "roc_auc_score": y_true = to_one_hot(y_true).astype(int) y_pred = y_pred[:, np.newaxis] else: y_true = y_true.astype(int) y_pred = from_one_hot(y_pred) #y_pred = from_one_hot(y_pred[:, np.newaxis]) print("y_pred") print(y_pred) if self.threshold is not None: y_pred = np.greater(y_pred, threshold) try: Loading
deepchem/models/__init__.py +4 −0 Original line number Diff line number Diff line Loading @@ -138,6 +138,10 @@ class Model(object): y_preds = [] batch_size = self.model_params["batch_size"] for (X_batch, y_batch, w_batch, ids_batch) in dataset.iterbatches(batch_size): print("y_batch.shape") print(y_batch.shape) print("self.predict_on_batch(X_batch).shape") print(self.predict_on_batch(X_batch).shape) y_pred_batch = np.reshape(self.predict_on_batch(X_batch), y_batch.shape) y_pred_batch = undo_transforms(y_pred_batch, transformers) y_preds.append(y_pred_batch) Loading
deepchem/models/keras_models/fcnet.py +17 −0 Original line number Diff line number Diff line Loading @@ -133,6 +133,23 @@ class MultiTaskDNN(KerasModel): y_pred = np.squeeze(y_pred) return y_pred def predict_proba_on_batch(self, X, n_classes=2): """ Makes predictions on given batch of new data. """ data = self.get_data_dict(X) y_pred_dict = self.raw_model.predict_on_batch(data) nb_samples = np.shape(X)[0] nb_tasks = len(self.tasks) y_pred = np.zeros((nb_samples, n_classes*nb_tasks)) for ind, task in enumerate(self.tasks): task_type = self.task_types[task] taskname = "task%d" % ind y_pred_task = np.squeeze(y_pred_dict[taskname]) y_pred[:, ind:ind+n_classes] = y_pred_task y_pred = np.squeeze(y_pred) return y_pred class SingleTaskDNN(MultiTaskDNN): """ Abstract base class for different ML models. Loading
deepchem/models/tensorflow_models/__init__.py +38 −16 Original line number Diff line number Diff line Loading @@ -95,6 +95,7 @@ class TensorflowGraph(object): This function constructs the computational graph for the model. It relies subclassed methods (build/cost) to construct specific graphs. """ print("TensorflowGraph.__init__()") self.graph = tf.Graph() self.model_params = model_params self.logdir = logdir Loading Loading @@ -136,6 +137,7 @@ class TensorflowGraph(object): def setup(self): """Add ops common to training/eval to the graph.""" print("TensorflowGraph.setup()") with self.graph.as_default(): with tf.name_scope('core_model'): self.build() Loading @@ -160,13 +162,15 @@ class TensorflowGraph(object): return tf.name_scope(self._name_scopes[name]) def add_training_cost(self): print("TensorflowGraph.add_training_cost()") print("self.output") print(self.output) with self.graph.as_default(): self.require_attributes(['output', 'labels', 'weights']) epsilon = 1e-3 # small float to avoid dividing by zero model_params = self.model_params weighted_costs = [] # weighted costs for each example gradient_costs = [] # costs used for gradient calculation old_costs = [] # old-style cost with self._shared_name_scope('costs'): for task in xrange(self.num_tasks): Loading @@ -182,28 +186,23 @@ class TensorflowGraph(object): # non-zero weight examples in the batch. Also, instead of using # tf.reduce_mean (which can put ops on the CPU) we explicitly # calculate with div/sum so it stays on the GPU. print("model_params['batch_size']") print(model_params['batch_size']) print("weighted_cost") print(weighted_cost) gradient_cost = tf.div(tf.reduce_sum(weighted_cost), model_params["batch_size"]) gradient_costs.append(gradient_cost) with tf.name_scope('old_cost'): old_cost = tf.div( tf.reduce_sum(weighted_cost), tf.reduce_sum(self.weights[task]) + epsilon) old_costs.append(old_cost) # aggregated costs with self._shared_name_scope('aggregated'): with tf.name_scope('gradient'): loss = tf.add_n(gradient_costs) with tf.name_scope('old_cost'): old_loss = tf.add_n(old_costs) # weight decay if model_params["penalty"] != 0.0: penalty = model_ops.WeightDecay(model_params) loss += penalty old_loss += penalty # loss used for gradient calculation self.loss = loss Loading Loading @@ -255,9 +254,22 @@ class TensorflowGraph(object): for (X_b, y_b, w_b, ids_b) in dataset.iterbatches(batch_size): # Run training op and compute summaries. feed_dict = self.construct_feed_dict(X_b, y_b, w_b, ids_b) step, loss, _ = sess.run( [train_op.values()[0], self.loss, self.updates], #step, loss, _ = sess.run( # [train_op.values()[0], self.loss, self.updates], # feed_dict=feed_dict) output, step, loss, _ = sess.run( self.output + [train_op.values()[0], self.loss, self.updates], feed_dict=feed_dict) #print("loss") #print(loss) y_pred = np.squeeze(np.array(output)) #print("y_pred") #print(y_pred) y_b = y_b.flatten() #print("y_b") #print(y_b) #print(".5*np.sum((y_b - y_pred)**2)/len(y_b)") #print(.5*np.sum((y_b - y_pred)**2)/len(y_b)) # Save model checkpoints at end of epoch saver.save(sess, self._save_path, global_step=self.global_step) log('Ending epoch %d: loss %g' % (epoch, loss), self.verbosity) Loading Loading @@ -333,10 +345,13 @@ class TensorflowGraph(object): #labels = np.array(from_one_hot( # np.squeeze(np.concatenate(labels)), axis=-1)) labels = np.squeeze(np.concatenate(labels)) labels = np.array(labels)[:, 1] #labels = np.squeeze(np.concatenate(labels)) outputs = np.array(from_one_hot( np.squeeze(np.concatenate(output)), axis=-1)) #labels = np.array(labels)[:, 1] return np.copy(labels) #return np.copy(labels) return np.copy(outputs) def add_output_ops(self): """Replace logits with softmax outputs.""" Loading Loading @@ -658,7 +673,7 @@ class TensorflowRegressor(TensorflowGraph): A tensor with shape batch_size containing the weighted cost for each example. """ #return tf.mul(0.5 * tf.square(output - labels), weights) return tf.mul(0.5 * tf.square(output - labels), weights) def example_counts(self, y_true): """Get counts of examples in each class. Loading Loading @@ -734,6 +749,13 @@ class TensorflowModel(Model): """ return self.eval_model.predict_on_batch(X) def predict_proba_on_batch(self, X): """ Makes predictions on batch of data. """ return self.eval_model.predict_proba_on_batch(X) def save(self): """ No-op since tf models save themselves during fit() Loading
deepchem/models/tensorflow_models/fcnet.py +117 −36 Original line number Diff line number Diff line Loading @@ -62,11 +62,22 @@ import numpy as np import tensorflow as tf from tensorflow.python.platform import logging from deepchem.metrics import from_one_hot from deepchem.models.tensorflow_models import TensorflowClassifier from deepchem.models.tensorflow_models import TensorflowRegressor from deepchem.models.tensorflow_models import model_ops from deepchem.metrics import to_one_hot def softmax(x): """Simple numpy softmax implementation """ tmp = np.max(x, axis = 1) x -= tmp.reshape((x.shape[0], 1)) x = np.exp(x) tmp = np.sum(x, axis = 1) x /= tmp.reshape((x.shape[0], 1)) return x class TensorflowMultiTaskClassifier(TensorflowClassifier): """Implements an icml model as configured in a model_config.proto.""" Loading Loading @@ -148,6 +159,92 @@ class TensorflowMultiTaskClassifier(TensorflowClassifier): orig_dict["valid"] = np.ones((self.model_params["batch_size"],), dtype=bool) return self._get_feed_dict(orig_dict) def predict_proba_on_batch(self, X): """Return model output for the provided input. Restore(checkpoint) must have previously been called on this object. Args: dataset: deepchem.datasets.dataset object. Returns: Tuple of three numpy arrays with shape num_examples x num_tasks (x ...): output: Model outputs. labels: True labels. weights: Example weights. Note that the output and labels arrays may be more than 2D, e.g. for classifier models that return class probabilities. Raises: AssertionError: If model is not in evaluation mode. ValueError: If output and labels are not both 3D or both 2D. """ if not self._restored_model: self.restore() with self.graph.as_default(): assert not model_ops.is_training() self.require_attributes(['output', 'labels', 'weights']) # run eval data through the model num_tasks = self.num_tasks outputs, labels, weights = [], [], [] start = time.time() with self._get_shared_session().as_default(): batch_count = -1.0 feed_dict = self.construct_feed_dict(X) batch_start = time.time() batch_count += 1 data = self._get_shared_session().run( self.output + self.labels + self.weights, feed_dict=feed_dict) batch_outputs = np.asarray(data[:num_tasks], dtype=float) batch_labels = np.asarray(data[num_tasks:num_tasks * 2], dtype=float) batch_weights = np.asarray(data[num_tasks * 2:num_tasks * 3], dtype=float) # reshape to batch_size x num_tasks x ... if batch_outputs.ndim == 3 and batch_labels.ndim == 3: batch_outputs = batch_outputs.transpose((1, 0, 2)) batch_labels = batch_labels.transpose((1, 0, 2)) elif batch_outputs.ndim == 2 and batch_labels.ndim == 2: batch_outputs = batch_outputs.transpose((1, 0)) batch_labels = batch_labels.transpose((1, 0)) else: raise ValueError( 'Unrecognized rank combination for output and labels: %s %s' % (batch_outputs.shape, batch_labels.shape)) batch_weights = batch_weights.transpose((1, 0)) valid = feed_dict[self.valid.name] # only take valid outputs if np.count_nonzero(~valid): batch_outputs = batch_outputs[valid] batch_labels = batch_labels[valid] batch_weights = batch_weights[valid] outputs.append(batch_outputs) labels.append(batch_labels) weights.append(batch_weights) print("predict_proba_on_batch()") print("batch_outputs") print(batch_outputs) print("batch_labels") print(batch_labels) logging.info('Eval batch took %g seconds', time.time() - start) #labels = np.array(from_one_hot( # np.squeeze(np.concatenate(labels)), axis=-1)) ##labels = np.squeeze(np.concatenate(labels)) #labels = np.array(labels)[:, 1] # We apply softmax to predictions to get class probabilities. #outputs = np.array(from_one_hot(softmax(np.squeeze(np.concatenate(outputs))))) outputs = softmax(np.squeeze(np.concatenate(outputs))) print("outputs") print(outputs) #return np.copy(labels) return np.copy(outputs) class TensorflowMultiTaskRegressor(TensorflowRegressor): """Implements an icml model as configured in a model_config.proto.""" Loading @@ -158,6 +255,7 @@ class TensorflowMultiTaskRegressor(TensorflowRegressor): mol_features: Molecule descriptor (e.g. fingerprint) tensor with shape batch_size x num_features. """ print("ENTERING TensorflowMultiTaskRegressor.build") assert len(self.model_params["data_shape"]) == 1 num_features = self.model_params["data_shape"][0] with self.graph.as_default(): Loading Loading @@ -197,14 +295,17 @@ class TensorflowMultiTaskRegressor(TensorflowRegressor): prev_layer = layer prev_layer_size = layer_sizes[i] self.output = [tf.squeeze(model_ops.FullyConnectedLayer( self.output = [] for task in range(self.num_tasks): self.output.append(tf.squeeze( model_ops.FullyConnectedLayer( tensor=prev_layer, size=layer_sizes[i], weight_init=tf.truncated_normal( shape=[prev_layer_size, 1], stddev=weight_init_stddevs[i]), bias_init=tf.constant(value=bias_init_consts[i], shape=[1])))] shape=[1])))) def construct_feed_dict(self, X_b, y_b=None, w_b=None, ids_b=None): """Construct a feed dictionary from minibatch data. Loading Loading @@ -259,52 +360,32 @@ class TensorflowMultiTaskRegressor(TensorflowRegressor): self.restore() with self.graph.as_default(): assert not model_ops.is_training() self.require_attributes(['output', 'labels', 'weights']) self.require_attributes(['output']) # run eval data through the model num_tasks = self.num_tasks output, labels, weights = [], [], [] start = time.time() outputs = [] with self._get_shared_session().as_default(): batch_count = -1.0 feed_dict = self.construct_feed_dict(X) batch_start = time.time() batch_count += 1 data = self._get_shared_session().run( self.output + self.labels + self.weights, feed_dict=feed_dict) batch_output = np.asarray(data[:num_tasks], dtype=float) batch_labels = np.asarray(data[num_tasks:num_tasks * 2], dtype=float) batch_weights = np.asarray(data[num_tasks * 2:num_tasks * 3], dtype=float) self.output, feed_dict=feed_dict) batch_outputs = np.asarray(data[:num_tasks], dtype=float) # reshape to batch_size x num_tasks x ... if batch_output.ndim == 3 and batch_labels.ndim == 3: batch_output = batch_output.transpose((1, 0, 2)) batch_labels = batch_labels.transpose((1, 0, 2)) elif batch_output.ndim == 2 and batch_labels.ndim == 2: batch_output = batch_output.transpose((1, 0)) batch_labels = batch_labels.transpose((1, 0)) if batch_outputs.ndim == 3: batch_outputs = batch_outputs.transpose((1, 0, 2)) elif batch_outputs.ndim == 2: batch_outputs = batch_outputs.transpose((1, 0)) else: raise ValueError( 'Unrecognized rank combination for output and labels: %s %s' % (batch_output.shape, batch_labels.shape)) batch_weights = batch_weights.transpose((1, 0)) 'Unrecognized rank combination for output: %s' % (batch_outputs.shape)) valid = feed_dict[self.valid.name] # only take valid outputs if np.count_nonzero(~valid): batch_output = batch_output[valid] batch_labels = batch_labels[valid] batch_weights = batch_weights[valid] output.append(batch_output) labels.append(batch_labels) weights.append(batch_weights) logging.info('Eval batch took %g seconds', time.time() - start) batch_outputs = batch_outputs[valid] outputs.append(batch_outputs) #labels = np.array(from_one_hot( # np.squeeze(np.concatenate(labels)), axis=-1)) labels = np.squeeze(np.concatenate(labels)) outputs = np.squeeze(np.concatenate(outputs)) return np.copy(labels) return np.copy(outputs)
