# I am unable to increase model accuracy

**URL:** <https://community.deeplearning.ai/t/i-am-unable-to-increase-model-accuracy/257460>\
**Category:** Natural Language Processing in TensorFlow\
**Tags:** week-module-4\
**Created:** [January 15, 2023, 4:11pm UTC](https://community.deeplearning.ai/t/i-am-unable-to-increase-model-accuracy/257460 "2023-01-15T16:11:12Z")\
**Posts on this page:** 5\
**Page:** 1

<div class="post-metadata">

**Author:** ![S.Dheeraj](https://avatars.discourse-cdn.com/v4/letter/s/5fc32e/32.png) [@S.Dheeraj](https://community.deeplearning.ai/u/S.Dheeraj)\
**Post date:** [January 15, 2023, 4:11pm UTC](https://community.deeplearning.ai/t/i-am-unable-to-increase-model-accuracy/257460/1 "2023-01-15T16:11:12Z")

</div>

Even though I used different layer’s in Bidirectional(LSTM()),I went till 2048 but maximum accuracy I obtained is Ranging between 60% to 67%

---

<div class="post-metadata">

**Author:** ![balaji.ambresh](https://sea1.discourse-cdn.com/dlai/user_avatar/community.deeplearning.ai/balaji.ambresh/32/128697_2.png) [@balaji.ambresh](https://community.deeplearning.ai/u/balaji.ambresh)\
**Post date:** [January 15, 2023, 5:19pm UTC](https://community.deeplearning.ai/t/i-am-unable-to-increase-model-accuracy/257460/2 "2023-01-15T17:19:39Z")

</div>

Please share the following details:

1. The values of units did use for LSTM.
2. Size of the embedding dimension.

---

<div class="post-metadata">

**Author:** ![S.Dheeraj](https://avatars.discourse-cdn.com/v4/letter/s/5fc32e/32.png) [@S.Dheeraj](https://community.deeplearning.ai/u/S.Dheeraj)\
**Post date:** [January 15, 2023, 5:45pm UTC](https://community.deeplearning.ai/t/i-am-unable-to-increase-model-accuracy/257460/3 "2023-01-15T17:45:19Z")

</div>

Sir, I used 128,256,512,1024,2048 and the Embedding dimensions are input\_dim=total\_words,output\_dim=100, input\_length=max\_sequence\_len-1

---

<div class="post-metadata">

**Author:** ![balaji.ambresh](https://sea1.discourse-cdn.com/dlai/user_avatar/community.deeplearning.ai/balaji.ambresh/32/128697_2.png) [@balaji.ambresh](https://community.deeplearning.ai/u/balaji.ambresh)\
**Post date:** [January 16, 2023, 5:59am UTC](https://community.deeplearning.ai/t/i-am-unable-to-increase-model-accuracy/257460/4 "2023-01-16T05:59:39Z")

</div>

Please click my name and message your notebook as an attachment.

---

<div class="post-metadata">

**Author:** ![balaji.ambresh](https://sea1.discourse-cdn.com/dlai/user_avatar/community.deeplearning.ai/balaji.ambresh/32/128697_2.png) [@balaji.ambresh](https://community.deeplearning.ai/u/balaji.ambresh)\
**Post date:** [January 16, 2023, 9:04am UTC](https://community.deeplearning.ai/t/i-am-unable-to-increase-model-accuracy/257460/5 "2023-01-16T09:04:26Z")

</div>

The learning rate of `1e-2` is high for the optimizer for your training setup. A higher learning rate does not mean that model weights will converge quicker. While it’s true that weight updates will involve larger steps, the performance is affected by oscillations around optimal weights.

The default learning rate of [rmsprop](https://www.tensorflow.org/api_docs/python/tf/keras/optimizers/experimental/RMSprop) is `1e-3`. When tuning learning rate, it’s important to experiment with candidate values and see how the model loss changes for different values. When in doubt, start with the default learning rate and tune your model architecture.

Try even other optimizers like `adam` to train your model as well.