Model architecture --->
adam=keras.optimizers.Adam(lr=0.0001)
early_stopping = keras.callbacks.ModelCheckpoint(filepath, monitor='val_acc', verbose=0, save_best_only=True, save_weights_only=False, mode='auto', period=10)
input_text = Input(shape=(1,), dtype="string")
embedding = Lambda(UniversalEmbedding, output_shape=(512, ))(input_text)
dense = Dense(1024)(embedding)
bnorm = BatchNormalization()(dense)
acti = Activation('relu')(bnorm)
pred = Dense(number_classes, activation='softmax')(acti)
model = Model(inputs=[input_text], outputs=pred)
model.compile(loss='categorical_crossentropy', optimizer=adam, metrics=['accuracy',f1score,sensitivity,precision])

I loaded entire model as --->
with open('universal_model.json', 'r') as f:
model = model_from_json(f.read())
model.load_weights('./universal_model.h5')

Now fro prediction -->
predicts = model.predict(np.array(["hi"], dtype=object))
I get this error --->
FailedPreconditionError (see above for traceback): Attempting to use uninitialized value
module/Encoder_en/DNN/ResidualHidden_3/projection
[[{{node module_apply_default/Encoder_en/DNN/ResidualHidden_3/projection/read}} =
Identity[T=DT_FLOAT, _device="/job:localhost/replica:0/task:0/device:CPU:0"] .
(module/Encoder_en/DNN/ResidualHidden_3/projection)]]

TF version -- 1.11.0
Keras -- 2.2.4

After get dataframe function cell,
df_train = get_dataframe('test_data.txt') has been inserted.
And model got trained on the same.
500 samples instead of 5452 samples.

There is a cell to read TEST data into the TRAINNG dataframe,
"df_train = get_dataframe('test_data.txt')"
right after reading the training data into the same dataframe, which makes 'train_text.shape' (500, 1) instead of (5452, 1).
I guess that should be removed.