# Copyright 2020 Huawei Technologies Co., Ltd # # Licensed under the Apache License, Version 2.0 (the "License"); # you may not use this file except in compliance with the License. # You may obtain a copy of the License at # # http://www.apache.org/licenses/LICENSE-2.0 # # Unless required by applicable law or agreed to in writing, software # distributed under the License is distributed on an "AS IS" BASIS, # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. # See the License for the specific language governing permissions and # limitations under the License. # ============================================================================ """ test_training """ import os import numpy as np from sklearn.metrics import roc_auc_score import mindspore.common.dtype as mstype from mindspore.ops import functional as F from mindspore.ops import composite as C from mindspore.ops import operations as P from mindspore.nn import Dropout from mindspore.nn.optim import Adam from mindspore.nn.metrics import Metric from mindspore import nn, ParameterTuple, Parameter from mindspore.common.initializer import Uniform, initializer, Normal from mindspore.train.callback import ModelCheckpoint, CheckpointConfig from .callback import EvalCallBack, LossCallBack np_type = np.float32 ms_type = mstype.float32 class AUCMetric(Metric): """AUC metric for DeepFM model.""" def __init__(self): super(AUCMetric, self).__init__() self.pred_probs = [] self.true_labels = [] def clear(self): """Clear the internal evaluation result.""" self.pred_probs = [] self.true_labels = [] def update(self, *inputs): batch_predict = inputs[1].asnumpy() batch_label = inputs[2].asnumpy() self.pred_probs.extend(batch_predict.flatten().tolist()) self.true_labels.extend(batch_label.flatten().tolist()) def eval(self): if len(self.true_labels) != len(self.pred_probs): raise RuntimeError('true_labels.size() is not equal to pred_probs.size()') auc = roc_auc_score(self.true_labels, self.pred_probs) return auc def init_method(method, shape, name, max_val=0.01): """ The method of init parameters. Args: method (str): The method uses to initialize parameter. shape (list): The shape of parameter. name (str): The name of parameter. max_val (float): Max value in parameter when uses 'random' or 'uniform' to initialize parameter. Returns: Parameter. """ if method in ['random', 'uniform']: params = Parameter(initializer(Uniform(max_val), shape, ms_type), name=name) elif method == "one": params = Parameter(initializer("ones", shape, ms_type), name=name) elif method == 'zero': params = Parameter(initializer("zeros", shape, ms_type), name=name) elif method == "normal": params = Parameter(initializer(Normal(max_val), shape, ms_type), name=name) return params def init_var_dict(init_args, values): """ Init parameter. Args: init_args (list): Define max and min value of parameters. values (list): Define name, shape and init method of parameters. Returns: dict, a dict ot Parameter. """ var_map = {} _, _max_val = init_args for key, shape, init_flag in values: if key not in var_map.keys(): if init_flag in ['random', 'uniform']: var_map[key] = Parameter(initializer(Uniform(_max_val), shape, ms_type), name=key) elif init_flag == "one": var_map[key] = Parameter(initializer("ones", shape, ms_type), name=key) elif init_flag == "zero": var_map[key] = Parameter(initializer("zeros", shape, ms_type), name=key) elif init_flag == 'normal': var_map[key] = Parameter(initializer(Normal(_max_val), shape, ms_type), name=key) return var_map class DenseLayer(nn.Cell): """ Dense Layer for Deep Layer of DeepFM Model; Containing: activation, matmul, bias_add; Args: input_dim (int): the shape of weight at 0-aixs; output_dim (int): the shape of weight at 1-aixs, and shape of bias weight_bias_init (list): weight and bias init method, "random", "uniform", "one", "zero", "normal"; act_str (str): activation function method, "relu", "sigmoid", "tanh"; keep_prob (float): Dropout Layer keep_prob_rate; scale_coef (float): input scale coefficient; """ def __init__(self, input_dim, output_dim, weight_bias_init, act_str, keep_prob=0.9, scale_coef=1.0): super(DenseLayer, self).__init__() weight_init, bias_init = weight_bias_init self.weight = init_method(weight_init, [input_dim, output_dim], name="weight") self.bias = init_method(bias_init, [output_dim], name="bias") self.act_func = self._init_activation(act_str) self.matmul = P.MatMul(transpose_b=False) self.bias_add = P.BiasAdd() self.cast = P.Cast() self.dropout = Dropout(keep_prob=keep_prob) self.mul = P.Mul() self.realDiv = P.RealDiv() self.scale_coef = scale_coef def _init_activation(self, act_str): act_str = act_str.lower() if act_str == "relu": act_func = P.ReLU() elif act_str == "sigmoid": act_func = P.Sigmoid() elif act_str == "tanh": act_func = P.Tanh() return act_func def construct(self, x): """Dense Layer for Deep Layer of DeepFM Model.""" x = self.act_func(x) if self.training: x = self.dropout(x) x = self.mul(x, self.scale_coef) x = self.cast(x, mstype.float16) weight = self.cast(self.weight, mstype.float16) wx = self.matmul(x, weight) wx = self.cast(wx, mstype.float32) wx = self.realDiv(wx, self.scale_coef) output = self.bias_add(wx, self.bias) return output class DeepFMModel(nn.Cell): """ From paper: "DeepFM: A Factorization-Machine based Neural Network for CTR Prediction" Args: batch_size (int): smaple_number of per step in training; (int, batch_size=128) filed_size (int): input filed number, or called id_feature number; (int, filed_size=39) vocab_size (int): id_feature vocab size, id dict size; (int, vocab_size=200000) emb_dim (int): id embedding vector dim, id mapped to embedding vector; (int, emb_dim=100) deep_layer_args (list): Deep Layer args, layer_dim_list, layer_activator; (int, deep_layer_args=[[100, 100, 100], "relu"]) init_args (list): init args for Parameter init; (list, init_args=[min, max, seeds]) weight_bias_init (list): weight, bias init method for deep layers; (list[str], weight_bias_init=['random', 'zero']) keep_prob (float): if dropout_flag is True, keep_prob rate to keep connect; (float, keep_prob=0.8) """ def __init__(self, config): super(DeepFMModel, self).__init__() self.batch_size = config.batch_size self.field_size = config.data_field_size self.vocab_size = config.data_vocab_size self.emb_dim = config.data_emb_dim self.deep_layer_dims_list, self.deep_layer_act = config.deep_layer_args self.init_args = config.init_args self.weight_bias_init = config.weight_bias_init self.keep_prob = config.keep_prob init_acts = [('W_l2', [self.vocab_size, 1], 'normal'), ('V_l2', [self.vocab_size, self.emb_dim], 'normal'), ('b', [1], 'normal')] var_map = init_var_dict(self.init_args, init_acts) self.fm_w = var_map["W_l2"] self.fm_b = var_map["b"] self.embedding_table = var_map["V_l2"] # Deep Layers self.deep_input_dims = self.field_size * self.emb_dim + 1 self.all_dim_list = [self.deep_input_dims] + self.deep_layer_dims_list + [1] self.dense_layer_1 = DenseLayer(self.all_dim_list[0], self.all_dim_list[1], self.weight_bias_init, self.deep_layer_act, self.keep_prob) self.dense_layer_2 = DenseLayer(self.all_dim_list[1], self.all_dim_list[2], self.weight_bias_init, self.deep_layer_act, self.keep_prob) self.dense_layer_3 = DenseLayer(self.all_dim_list[2], self.all_dim_list[3], self.weight_bias_init, self.deep_layer_act, self.keep_prob) self.dense_layer_4 = DenseLayer(self.all_dim_list[3], self.all_dim_list[4], self.weight_bias_init, self.deep_layer_act, self.keep_prob) # FM, linear Layers self.Gatherv2 = P.GatherV2() self.Mul = P.Mul() self.ReduceSum = P.ReduceSum(keep_dims=False) self.Reshape = P.Reshape() self.Square = P.Square() self.Shape = P.Shape() self.Tile = P.Tile() self.Concat = P.Concat(axis=1) self.Cast = P.Cast() def construct(self, id_hldr, wt_hldr): """ Args: id_hldr: batch ids; [bs, field_size] wt_hldr: batch weights; [bs, field_size] """ mask = self.Reshape(wt_hldr, (self.batch_size, self.field_size, 1)) # Linear layer fm_id_weight = self.Gatherv2(self.fm_w, id_hldr, 0) wx = self.Mul(fm_id_weight, mask) linear_out = self.ReduceSum(wx, 1) # FM layer fm_id_embs = self.Gatherv2(self.embedding_table, id_hldr, 0) vx = self.Mul(fm_id_embs, mask) v1 = self.ReduceSum(vx, 1) v1 = self.Square(v1) v2 = self.Square(vx) v2 = self.ReduceSum(v2, 1) fm_out = 0.5 * self.ReduceSum(v1 - v2, 1) fm_out = self.Reshape(fm_out, (-1, 1)) # Deep layer b = self.Reshape(self.fm_b, (1, 1)) b = self.Tile(b, (self.batch_size, 1)) deep_in = self.Reshape(vx, (-1, self.field_size * self.emb_dim)) deep_in = self.Concat((deep_in, b)) deep_in = self.dense_layer_1(deep_in) deep_in = self.dense_layer_2(deep_in) deep_in = self.dense_layer_3(deep_in) deep_out = self.dense_layer_4(deep_in) out = linear_out + fm_out + deep_out return out, fm_id_weight, fm_id_embs class NetWithLossClass(nn.Cell): """ NetWithLossClass definition. """ def __init__(self, network, l2_coef=1e-6): super(NetWithLossClass, self).__init__(auto_prefix=False) self.loss = P.SigmoidCrossEntropyWithLogits() self.network = network self.l2_coef = l2_coef self.Square = P.Square() self.ReduceMean_false = P.ReduceMean(keep_dims=False) self.ReduceSum_false = P.ReduceSum(keep_dims=False) def construct(self, batch_ids, batch_wts, label): predict, fm_id_weight, fm_id_embs = self.network(batch_ids, batch_wts) log_loss = self.loss(predict, label) mean_log_loss = self.ReduceMean_false(log_loss) l2_loss_w = self.ReduceSum_false(self.Square(fm_id_weight)) l2_loss_v = self.ReduceSum_false(self.Square(fm_id_embs)) l2_loss_all = self.l2_coef * (l2_loss_v + l2_loss_w) * 0.5 loss = mean_log_loss + l2_loss_all return loss class TrainStepWrap(nn.Cell): """ TrainStepWrap definition """ def __init__(self, network, lr=5e-8, eps=1e-8, loss_scale=1000.0): super(TrainStepWrap, self).__init__(auto_prefix=False) self.network = network self.network.set_train() self.weights = ParameterTuple(network.trainable_params()) self.optimizer = Adam(self.weights, learning_rate=lr, eps=eps, loss_scale=loss_scale) self.hyper_map = C.HyperMap() self.grad = C.GradOperation('grad', get_by_list=True, sens_param=True) self.sens = loss_scale def construct(self, batch_ids, batch_wts, label): weights = self.weights loss = self.network(batch_ids, batch_wts, label) sens = P.Fill()(P.DType()(loss), P.Shape()(loss), self.sens) # grads = self.grad(self.network, weights)(batch_ids, batch_wts, label, sens) return F.depend(loss, self.optimizer(grads)) class PredictWithSigmoid(nn.Cell): """ Eval model with sigmoid. """ def __init__(self, network): super(PredictWithSigmoid, self).__init__(auto_prefix=False) self.network = network self.sigmoid = P.Sigmoid() def construct(self, batch_ids, batch_wts, labels): logits, _, _, = self.network(batch_ids, batch_wts) pred_probs = self.sigmoid(logits) return logits, pred_probs, labels class ModelBuilder: """ Model builder for DeepFM. Args: model_config (ModelConfig): Model configuration. train_config (TrainConfig): Train configuration. """ def __init__(self, model_config, train_config): self.model_config = model_config self.train_config = train_config def get_callback_list(self, model=None, eval_dataset=None): """ Get callbacks which contains checkpoint callback, eval callback and loss callback. Args: model (Cell): The network is added callback (default=None). eval_dataset (Dataset): Dataset for eval (default=None). """ callback_list = [] if self.train_config.save_checkpoint: config_ck = CheckpointConfig(save_checkpoint_steps=self.train_config.save_checkpoint_steps, keep_checkpoint_max=self.train_config.keep_checkpoint_max) ckpt_cb = ModelCheckpoint(prefix=self.train_config.ckpt_file_name_prefix, directory=self.train_config.output_path, config=config_ck) callback_list.append(ckpt_cb) if self.train_config.eval_callback: if model is None: raise RuntimeError("train_config.eval_callback is {}; get_callback_list() args model is {}".format( self.train_config.eval_callback, model)) if eval_dataset is None: raise RuntimeError("train_config.eval_callback is {}; get_callback_list() " "args eval_dataset is {}".format(self.train_config.eval_callback, eval_dataset)) auc_metric = AUCMetric() eval_callback = EvalCallBack(model, eval_dataset, auc_metric, eval_file_path=os.path.join(self.train_config.output_path, self.train_config.eval_file_name)) callback_list.append(eval_callback) if self.train_config.loss_callback: loss_callback = LossCallBack(loss_file_path=os.path.join(self.train_config.output_path, self.train_config.loss_file_name)) callback_list.append(loss_callback) if callback_list: return callback_list return None def get_train_eval_net(self): deepfm_net = DeepFMModel(self.model_config) loss_net = NetWithLossClass(deepfm_net, l2_coef=self.train_config.l2_coef) train_net = TrainStepWrap(loss_net, lr=self.train_config.learning_rate, eps=self.train_config.epsilon, loss_scale=self.train_config.loss_scale) eval_net = PredictWithSigmoid(deepfm_net) return train_net, eval_net