tf.keras.losses.MeanSquaredError() is not available in lower versions of tensorflow. It seems that versions above 1.15 are useful on the official website. Anyway, I can’t use 1.7. Reinstalling tensorflow1.15 will report an error again, so I found a Alternative method def compute_loss(labels, logits): return tf.reduce_mean(tf.keras.losses.mse(labels, logits)) loss = compute_loss(y, logits) can be used
instead of ComputeLoss = tf.keras.losses.MeanSquaredError() loss = ComputeLoss(y, logits)
用
def compute_metrics(labels, logits):
return tf.keras.metrics.mae(labels, logits)
ComputeMetrics = compute_metrics(y, logits)
代替
ComputeMetrics = tf.keras.metrics.MeanAbsoluteError()
ComputeMetrics(y, logits)
The v2 version of tf:
if tf.io.gfile.exists(MODEL_DIR):
pass
else: The v1 version corresponding to
tf.io.gfile.makedirs(MODEL_DIR) : (can be found when searching the v2 version on the official website) if tf.gfile .Exists(MODEL_DIR): pass else: tf.gfile.MakeDirs(MODEL_DIR)