预测特斯拉股票时遇NumPy转Tensor错误，求数据适配方案

Hey there, let's break down why you're seeing that ValueError: Failed to convert a NumPy array to a Tensor and get your model up and running. The issue boils down to a critical typo in your preprocessing code, plus a few misaligned model settings for your classification task.

1. The Root Cause: Incorrect Data Collection in Preprocessing #

Look at this line in your preproccesing function:

X.append(sequential_data)

You're appending the entire sequential_data list (which contains all sequences and targets) to X every time, instead of appending just the individual sequence seq. This creates a deeply nested NumPy array where each element is the full dataset—TensorFlow can't parse that into a valid input tensor.

Fix it by changing that line to:

X.append(seq)

2. Align Your Loss Function with the Task Type #

You're doing a binary classification task (predicting 0 or 1), but you're using mse (mean squared error)—a loss function designed for regression problems. This will lead to poor performance and potential training issues.

Update your model compilation to use a classification-friendly loss:
Since your train_y are integer labels (0/1) and your output layer uses softmax with 2 units, use sparse_categorical_crossentropy (no need to one-hot encode your labels):

model.compile(loss="sparse_categorical_crossentropy", optimizer=opt, metrics=["accuracy"])

3. Clean Up Redundant LSTM Input Shape Declarations #

Only the first LSTM layer needs the input_shape parameter—subsequent LSTM layers automatically inherit the input shape from the previous layer's output. Remove the input_shape from the second and third LSTM layers to avoid confusion:

model.add(LSTM(128, input_shape=(train_x.shape[1:]), activation="relu", return_sequences=True))
model.add(Dropout(0.2))
model.add(BatchNormalization())
model.add(LSTM(128, activation="relu", return_sequences=True))  # No input_shape needed here
model.add(Dropout(0.2))
model.add(BatchNormalization())
model.add(LSTM(128, activation="relu", return_sequences=True))  # Or here
model.add(Dropout(0.2))
model.add(BatchNormalization())

4. Verify Your Data Shapes (Post-Fix) #

After making the above changes, print your input shapes to confirm they match LSTM requirements (samples, time steps, features):

print(train_x.shape)  # Should be (number_of_samples, 100, 2)

This ensures TensorFlow can properly process your sequential data.

Full Corrected Preprocessing Function #

Here's the fixed preproccesing function with the critical typo fixed:

def preproccesing(data):
    data = data.drop("future", 1)
    for col in data.columns:
        if col != "target":
            data[col] = data[col].pct_change()
            data.dropna(inplace=True)
            data[col] = preprocessing.scale(data[col].values)
            data.dropna(inplace = True)
    sequential_data = []
    prev_days = deque(maxlen=Back_period_history)
    for i in data.values:
        prev_days.append([n for n in i[:-1]])
        if len(prev_days) == Back_period_history:
            sequential_data.append([np.array(prev_days), i[-1]])
    rd.shuffle(sequential_data)
    buys = []
    sells = []
    for seq, target in sequential_data:
        if target == 0:
            sells.append([seq, target])
        elif target == 1:
            buys.append([seq,target])
    rd.shuffle(buys)
    rd.shuffle(sells)
    lower = min(len(buys), len(sells))
    buys = buys[:lower]
    sells = sells[:lower]
    sequential_data = buys+sells
    rd.shuffle(sequential_data)
    X = []
    y = []
    for seq, target in sequential_data:
        X.append(seq)  # Fixed line!
        y.append(target)
    return np.array(X),y

Once you apply these fixes, your model should train without the tensor conversion error and perform better on your classification task.

内容的提问来源于stack exchange，提问作者manan goyal