#!/usr/bin/env python
# coding: utf-8

# # Inspect
# 
# We show how to inspect results and intermediate results
# using the example pages in the *example* directory.
# 
# Reference: [page](https://among.github.io/fusus/fusus/page.html).

# Enable auto-loading of changed code.
# This is handy if you are a developer and changing `fusus` code wanting to load changed code on the fly.

# In[1]:


get_ipython().run_line_magic('load_ext', 'autoreload')
get_ipython().run_line_magic('autoreload', '2')


# In[2]:


from fusus.book import Book


# In[3]:


B = Book(cd="~/github/among/fusus/example")


# In[4]:


# cd to the book directory
get_ipython().system('cd `pwd`')


# In[5]:


get_ipython().system('echo `pwd`')


# In[6]:


B.availablePages()


# ## Inspection
# 
# We run a single page and switch batch mode off.

# In[7]:


page = B.process(batch=False, pages='47')


# ### Inspect intermediate stages
# 
# We get a handle to the page object that has been processed,
# and because batch mode is off, it has retained its intermediate results.
# 
# Let's show all stages, but heavily reduced:

# In[8]:


page.show(stage="proofword")


# In[9]:


page.show(width=200)


# Note the data stages:
# 
# *    `markData` with the cleaning data.
# *    `char` with the OCR data at character level.
# *    `word` with the OCR data at word level.
# *    `line` with the locations of the line boxes.
# 
# The proof stages (`proofword` and `proofchar`) are HTML pages that cannot easily be shown inside the notebook.
# It is better to go to the indicated file on your system and view them directly.

# Now the histogram and the `cleanh` stage at true size;

# In[10]:


page.show(stage='histogram,cleanh')


# The grey rectangles are the traces of wiped marks.
# 
# But what goes to the OCR engine is stage `clean`:

# In[11]:


page.show(stage='clean')


# ### Inspect bands
# 
# Let's have a look at all the bands.
# 
# We project the bands on the histogram of the page.

# In[12]:


for doBands in ("high", "low", "mid", "main", "broad", "inter"):
    page.show(stage="histogram", band=doBands)


# We can also show several bands together:

# In[13]:


page.show(stage="histogram", band="low,high")


# # Cleaning
# 
# Proper cleaning requires some interactive adjusting of parameters.
# So we need feedback as to what marks have been cleaned and why.

# ## Connectedness
# 
# We can use the `boxed` stage to inspect the results of cleaning.
# 
# The letters above the boxes refer to the bands in which those marks are searched:
# 
# letter | band
# --- | ---
# ` ` | main
# `l` | low
# `m` | mid
# `h` | high
# `b` | broad
# `i` | inter
# 
# The numbers above the boxes are the numbers of the marks, by running `B.availableMarks()` you see the marks with their numbers.
# 
# Marks in *green* boxes will not be removed. 
# They are apparently part of a letter, not an isolated blob of ink.
# 
# The algorithm computes a degree of connection, and if that is bigger than 0.1, the mark counts as connected.
# The connection degrees are printed below the marks (times 100), so 10 is the cutoff value. 
# Green marks are always higher than 10.
# 
# Marks in *orange* boxes will be removed. 
# They have a small connection degree, less than 0.1, but usually much smaller. If it is near zero, it will be left out.

# In[14]:


page.show(stage='boxed')


# We can also ask for a log of the cleaning process which gives a bit more information:

# In[15]:


page.show(stage='markData')


# The starred entries correspond to the green boxes.
# 
# The connectivity times 100 is the number under the boxes.

# Now we only want to see the effects of cleaning for specific marks (in specific bands):

# In[16]:


page.show(stage='boxed', band="high,mid", mark="comma,a")


# ## Cleaning parameters
# 
# The parameters that steer the cleaning process are
# 
# * `connectBorder`
# * `connectRatio`
# * `connectThreshold`
# 
# See the
# [settings specs](https://among.github.io/fusus/fusus/parameters.html).

# In[ ]: