ValueError: Input 0 of layer conv1d is incompatible with the layer: : expected min

So I was tinkering with some code for time series forecasting. I have dealt with this error before (the formatting of my data was wrong). But in this case I can't figure out what I've done wrong. Here is the source of the problem

    monk= tf.keras.models.Sequential()
    monk.add(tf.keras.layers.Flatten())
    monk.add(tf.keras.layers.Conv1D(64,2,input_shape=(X_train.shape[1],X_train.shape[2])))
    monk.add(tf.keras.layers.MaxPool1D())
    monk.add(tf.keras.layers.Activation('relu'))
    monk.add(tf.keras.layers.Dense(32))
    monk.add(tf.keras.layers.Dense(1,'sigmoid'))
    monk.compile('adam','binary_crossentropy',['accuracy'])
    monk.fit(X_train,y_train,epochs=10)

where the shape of X_train is (100,5,1) and the shape of y_train is (100,)

The fully reproducible code

from random import shuffle
from torch import are_deterministic_algorithms_enabled
import yfinance as yf
import tensorflow as tf
import datetime 
import time
import numpy as np
def retrain(symbol):

    todayy  = [int(item) for item in str(datetime.datetime.today()).split(' ')[0].split('-')]
    start = datetime.datetime(todayy[0]-2,todayy[1],todayy[2])
    end = datetime.datetime(todayy[0],todayy[1],todayy[2])
    stock = yf.download(symbol,start=start,end=end)
    print(stock)
    buy = []
    for x in range(stock.shape[0]):
        open = stock.iloc[x]['Open']
        close=stock.iloc[x]['Close']
        if close-open>0:
            buy.append(1)
        else:
            buy.append(0)
    print(buy)
    X = []
    y= []
    temp=[]
    for x in range(len(buy)):
        item = buy[x]
        temp.append(np.array([item]))
        if len(temp)>=5:
            X.append(np.array(temp))
            temp=[]
            try:
                y.append(buy[x 1])
            except: 
                break
    buyz=[]
    sellz=[]
    for item in list(zip(X,y)):
        print(item)
        if item[1]==1:
            buyz.append(item)
        else:
            sellz.append(item)
    
    buyz = buyz[:min(len(buyz),len(sellz))]
    selzz = sellz[:min(len(buyz),len(sellz))]
    
    all = []
    for item in buyz:
        all.append(item)
    for item in sellz:
        all.append(item)
    shuffle(all)
    X_train = []
    y_train =[]
    for item in all:
        print(item)
        X_train.append(item[0])
        y_train.append(item[1])
    #input()
    X_train=np.array(X_train)
    y_train=np.array(y_train)
    print(X_train)
    print(y_train)
    print(X_train.shape)
    print(y_train.shape)
    monk= tf.keras.models.Sequential()
    monk.add(tf.keras.layers.Flatten())
    monk.add(tf.keras.layers.Conv1D(64,2,input_shape=(X_train.shape[1],X_train.shape[2])))
    monk.add(tf.keras.layers.MaxPool1D())
    monk.add(tf.keras.layers.Activation('relu'))
    monk.add(tf.keras.layers.Dense(32))
    monk.add(tf.keras.layers.Dense(1,'sigmoid'))
    monk.compile('adam','binary_crossentropy',['accuracy'])
    monk.fit(X_train,y_train,epochs=10)
    #print(monk(X))
            
        
retrain('LEVI')

Any help would be much appreciated.

CodePudding user response：

Remove tf.keras.layers.Flatten(), since it is flattening your 3D tensor (batch size, timesteps, features) to (batch size, features).

You can should add the Flatten layer again after tf.keras.layers.Activation('relu').