support deepfm and kaggle display advertising dataset

requirements.txt

@@ -13,4 +13,6 @@ gensim==3.8.1
 PyYAML
 opencv-python==4.1.2.30
 flask_cors>=3.0.10
-pycocotools>=2.0.0          # for st test
+pycocotools>=2.0.0          # for st test
+wget==3.2
+scikit-learn==1.0.1

setup.py

@@ -55,6 +55,9 @@ def _write_version(file):
     'PyYAML',
     'opencv-python==4.1.2.30',
     'flask_cors>=3.0.10',
+    'wget==3.2',
+    'scikit-learn==1.0.1',
+    'tqdm==4.62.3',
 ]
 
 test_required_package = [

tinyms/callbacks.py

@@ -20,7 +20,7 @@
 from mindspore.train.callback import *
 from . import Tensor
 
-__all__ = ['LossTimeMonitor', 'BertLossCallBack']
+__all__ = ['LossTimeMonitor', 'LossTimeMonitorV2', 'BertLossCallBack']
 __all__.extend(callback.__all__)
 
 
@@ -81,6 +81,60 @@ def step_end(self, run_context):
             np.mean(self.losses), step_mseconds, self.lr_init[cb_params.cur_step_num - 1]))
 
 
+class LossTimeMonitorV2(Callback):
+    """
+    Monitor loss and time version 2.0.
+    This version will not show learning rate.
+
+    Args:
+
+    Returns:
+        None
+
+    Examples:
+        >>> from tinyms.callbacks import LossTimeMonitorV2
+        >>>
+        >>> LossTimeMonitorV2()
+    """
+
+    def __init__(self):
+        super(LossTimeMonitorV2, self).__init__()
+
+    def epoch_begin(self, run_context):
+        self.losses = []
+        self.epoch_time = time.time()
+
+    def epoch_end(self, run_context):
+        cb_params = run_context.original_args()
+
+        epoch_mseconds = (time.time() - self.epoch_time) * 1000
+        per_step_mseconds = epoch_mseconds / cb_params.batch_num
+        print("epoch time: {:5.3f}, per step time: {:5.3f}, avg loss: {:5.3f}".
+              format(epoch_mseconds, per_step_mseconds, np.mean(self.losses)), flush=True)
+
+    def step_begin(self, run_context):
+        self.step_time = time.time()
+
+    def step_end(self, run_context):
+        cb_params = run_context.original_args()
+        step_mseconds = (time.time() - self.step_time) * 1000
+        step_loss = cb_params.net_outputs
+        # arr_lr = cb_params.optimizer.learning_rate.asnumpy()
+        # lr = float(np.array2string(arr_lr))
+
+        if isinstance(step_loss, (tuple, list)) and isinstance(step_loss[0], Tensor):
+            step_loss = step_loss[0]
+        if isinstance(step_loss, Tensor):
+            step_loss = np.mean(step_loss.asnumpy())
+
+        self.losses.append(step_loss)
+        cur_step_in_epoch = (cb_params.cur_step_num - 1) % cb_params.batch_num
+
+        print("epoch: [{:3d}/{:3d}], step:[{:5d}/{:5d}], loss:[{:5.3f}/{:5.3f}], time:[{:5.3f}]]".format(
+            cb_params.cur_epoch_num - 1, cb_params.epoch_num, cur_step_in_epoch, cb_params.batch_num, step_loss,
+            np.mean(self.losses), step_mseconds), flush=True)
+
+
 class BertLossCallBack(Callback):
     """
     Monitor the loss in training.