Skip to content

mdlai/digit_recognition

Repository files navigation

Prerequisites and Instructions

The notebooks use the MNIST data and classify it with KNN and CNN.

To launch Tensorboard navigate to the tensor_flow_logs folder and use:

python ~/path/to/python/tensorflow/tensorboard/tensorboard.py --logdir=${PWD}

To launch the prediction app:

Run

python number_prediction_app.py

Open Browser at localhost:8080

About this Project

This repo is an app that uses a Convolutional Neural Network built in Tensorflow and trained on MNIST data to predict a number drawn by a user in d3.js

Fixes for new versions of TF

Error --> ModuleNotFoundError: No module named 'tensorflow.tensorboard.tensorboard' Solution --> tensorflow/tensorflow#10959

Run from: python ~/path/to/python/tensorflow/tensorboard/tensorboard.py --logdir=${PWD} to: tensorboard --logdir=output

Error --> 'module' object has no attribute 'scalar_summary' Solution --> https://stackoverflow.com/questions/41066244/tensorflow-module-object-has-no-attribute-scalar-summary

Change code from: tf.scalar_summary(...) to: tf.summary.scalar(...)

Error --> AttributeError: 'module' object has no attribute 'merge_all_summaries' Solution --> tensorflow/models#1066

Change code from: summary_op = tf.merge_all_summaries() to: summary_op = tf.summary.merge_all()

About

Hand Written Digit Predictor

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published