Codex/dockerize project

.dockerignore

@@ -0,0 +1,7 @@
+__pycache__/
+*.py[cod]
+*.so
+*.DS_Store
+.git
+.gitignore
+.env

.python-version

@@ -0,0 +1 @@
+3.9.18

Dockerfile

@@ -0,0 +1,22 @@
+# syntax=docker/dockerfile:1
+
+FROM python:3.11-slim
+
+ENV PYTHONDONTWRITEBYTECODE=1 \
+    PYTHONUNBUFFERED=1
+
+WORKDIR /app
+
+RUN apt-get update \
+    && apt-get install -y --no-install-recommends \
+        build-essential \
+    && rm -rf /var/lib/apt/lists/*
+
+COPY requirements.txt ./
+RUN pip install --no-cache-dir -r requirements.txt
+
+COPY . .
+
+EXPOSE 5000
+
+CMD ["python", "web_app.py", "--host", "0.0.0.0", "--port", "3000"]

demo.py

@@ -38,8 +38,19 @@ def __init__(self):
         )
         self.nn.restore()
 
-    def write(self, filename, lines, biases=None, styles=None, stroke_colors=None, stroke_widths=None):
+    def write(
+        self,
+        filename,
+        lines,
+        biases=None,
+        styles=None,
+        stroke_colors=None,
+        stroke_widths=None,
+        alignment="center",
+    ):
         valid_char_set = set(drawing.alphabet)
+        if alignment not in {"left", "center"}:
+            raise ValueError("Alignment must be either 'left' or 'center'.")
         for line_num, line in enumerate(lines):
             if len(line) > 75:
                 raise ValueError(
@@ -59,7 +70,14 @@ def write(self, filename, lines, biases=None, styles=None, stroke_colors=None, s
                     )
 
         strokes = self._sample(lines, biases=biases, styles=styles)
-        self._draw(strokes, lines, filename, stroke_colors=stroke_colors, stroke_widths=stroke_widths)
+        self._draw(
+            strokes,
+            lines,
+            filename,
+            stroke_colors=stroke_colors,
+            stroke_widths=stroke_widths,
+            alignment=alignment,
+        )
 
     def _sample(self, lines, biases=None, styles=None):
         num_samples = len(lines)
@@ -74,7 +92,7 @@ def _sample(self, lines, biases=None, styles=None):
         if styles is not None:
             for i, (cs, style) in enumerate(zip(lines, styles)):
                 x_p = np.load('styles/style-{}-strokes.npy'.format(style))
-                c_p = np.load('styles/style-{}-chars.npy'.format(style)).tostring().decode('utf-8')
+                c_p = np.load('styles/style-{}-chars.npy'.format(style)).tobytes().decode('utf-8')
 
                 c_p = str(c_p) + " " + cs
                 c_p = drawing.encode_ascii(c_p)
@@ -107,7 +125,15 @@ def _sample(self, lines, biases=None, styles=None):
         samples = [sample[~np.all(sample == 0.0, axis=1)] for sample in samples]
         return samples
 
-    def _draw(self, strokes, lines, filename, stroke_colors=None, stroke_widths=None):
+    def _draw(
+        self,
+        strokes,
+        lines,
+        filename,
+        stroke_colors=None,
+        stroke_widths=None,
+        alignment="center",
+    ):
         stroke_colors = stroke_colors or ['black']*len(lines)
         stroke_widths = stroke_widths or [2]*len(lines)
 
@@ -133,7 +159,11 @@ def _draw(self, strokes, lines, filename, stroke_colors=None, stroke_widths=None
 
             strokes[:, 1] *= -1
             strokes[:, :2] -= strokes[:, :2].min() + initial_coord
-            strokes[:, 0] += (view_width - strokes[:, 0].max()) / 2
+            if alignment == "center":
+                strokes[:, 0] += (view_width - strokes[:, 0].max()) / 2
+            else:
+                left_padding = 60
+                strokes[:, 0] += left_padding
 
             prev_eos = 1.0
             p = "M{},{} ".format(0, 0)

readme.md

@@ -31,8 +31,34 @@ hand.write(
 
 Currently, the `Hand` class must be imported from `demo.py`.  If someone would like to package this project to make it more usable, please [contribute](#contribute).
 
+### Web UI
+
+A lightweight Flask UI is available to generate SVG files directly from a browser. Install the dependencies and launch the
+development server:
+
+```
+pip install -r requirements.txt
+python web_app.py
+```
+
+Open <http://localhost:5000> and enter the text (multi-line is supported) you would like to render. Submitting the form downloads the generated SVG file.
+
+Use `python web_app.py --help` to see options for changing the host/port or enabling debug mode when you need Flask's auto
+reloader.
+
 A pretrained model is included, but if you'd like to train your own, read <a href='https://github.com/sjvasquez/handwriting-synthesis/tree/master/data/raw'>these instructions</a>.
 
+### Docker
+
+You can also run the web UI in a container:
+
+```bash
+docker build -t handwriting-synthesis .
+docker run --rm -p 5000:5000 handwriting-synthesis
+```
+
+Open <http://localhost:5000> after the container starts to use the interface.
+
 ## Demonstrations
 Below are a few hundred samples from the model, including some samples demonstrating the effect of priming and biasing the model.  Loosely speaking, biasing controls the neatness of the samples and priming controls the style of the samples. The code for these demonstrations can be found in `demo.py`.
 

requirements.txt

@@ -1,6 +1,10 @@
-matplotlib>=2.1.0
-pandas>= 0.22.0
-scikit-learn>=0.19.1
-scipy>=1.0.0
-svgwrite>=1.1.12
-tensorflow==1.6.0
+matplotlib>=3.8.4
+numpy>=1.26.4
+pandas>=2.2.2
+scikit-learn>=1.4.2
+scipy>=1.11.4
+svgwrite>=1.4.3
+tensorflow>=2.20.0
+tensorflow-probability>=0.25.0
+tf-keras>=2.20.0
+flask>=3.0.0

rnn.py

@@ -2,7 +2,7 @@
 import os
 
 import numpy as np
-import tensorflow as tf
+import tensorflow.compat.v1 as tf
 
 import drawing
 from data_frame import DataFrame
@@ -12,6 +12,9 @@
 from tf_utils import time_distributed_dense_layer
 
 
+tf.disable_v2_behavior()
+
+
 class DataReader(object):
 
     def __init__(self, data_dir):

rnn_cell.py

@@ -1,8 +1,10 @@
 from collections import namedtuple
 
-import tensorflow as tf
-import tensorflow.contrib.distributions as tfd
 import numpy as np
+import tensorflow.compat.v1 as tf
+import tensorflow_probability as tfp
+from tensorflow.python.ops.rnn_cell_impl import LSTMCell as TFCompatLSTMCell
+from tensorflow.python.ops.rnn_cell_impl import RNNCell as TFCompatRNNCell
 
 from tf_utils import dense_layer, shape
 
@@ -13,7 +15,12 @@
 )
 
 
-class LSTMAttentionCell(tf.nn.rnn_cell.RNNCell):
+tf.disable_v2_behavior()
+
+tfd = tfp.distributions
+
+
+class LSTMAttentionCell(TFCompatRNNCell):
 
     def __init__(
         self,
@@ -77,7 +84,7 @@ def __call__(self, inputs, state, scope=None):
 
             # lstm 1
             s1_in = tf.concat([state.w, inputs], axis=1)
-            cell1 = tf.contrib.rnn.LSTMCell(self.lstm_size)
+            cell1 = TFCompatLSTMCell(self.lstm_size)
             s1_out, s1_state = cell1(s1_in, state=(state.c1, state.h1))
 
             # attention
@@ -101,12 +108,12 @@ def __call__(self, inputs, state, scope=None):
 
             # lstm 2
             s2_in = tf.concat([inputs, s1_out, w], axis=1)
-            cell2 = tf.contrib.rnn.LSTMCell(self.lstm_size)
+            cell2 = TFCompatLSTMCell(self.lstm_size)
             s2_out, s2_state = cell2(s2_in, state=(state.c2, state.h2))
 
             # lstm 3
             s3_in = tf.concat([inputs, s2_out, w], axis=1)
-            cell3 = tf.contrib.rnn.LSTMCell(self.lstm_size)
+            cell3 = TFCompatLSTMCell(self.lstm_size)
             s3_out, s3_state = cell3(s3_in, state=(state.c3, state.h3))
 
             new_state = LSTMAttentionCellState(
@@ -155,7 +162,7 @@ def termination_condition(self, state):
         past_final_char = char_idx >= self.attention_values_lengths
         output = self.output_function(state)
         es = tf.cast(output[:, 2], tf.int32)
-        is_eos = tf.equal(es, np.ones_like(es))
+        is_eos = tf.equal(es, tf.ones_like(es))
         return tf.logical_or(tf.logical_and(final_char, is_eos), past_final_char)
 
     def _parse_parameters(self, gmm_params, eps=1e-8, sigma_eps=1e-4):

rnn_ops.py

@@ -1,17 +1,24 @@
+import tensorflow.compat.v1 as tf
 from tensorflow.python.framework import constant_op
 from tensorflow.python.framework import dtypes
 from tensorflow.python.framework import ops
 from tensorflow.python.ops import array_ops
-from tensorflow.python.ops import control_flow_ops
 from tensorflow.python.ops import math_ops
 from tensorflow.python.ops import tensor_array_ops
 from tensorflow.python.ops import variable_scope as vs
-from tensorflow.python.ops.rnn_cell_impl import _concat, _like_rnncell
+from tensorflow.python.ops.rnn_cell_impl import _concat
 from tensorflow.python.ops.rnn import _maybe_tensor_shape_from_tensor
 from tensorflow.python.util import nest
 from tensorflow.python.framework import tensor_shape
 from tensorflow.python.eager import context
 
+tf.disable_v2_behavior()
+
+
+def _is_rnn_cell(cell):
+    required_attrs = ("state_size", "output_size")
+    return all(hasattr(cell, attr) for attr in required_attrs) and callable(cell)
+
 
 def raw_rnn(cell, loop_fn, parallel_iterations=None, swap_memory=False, scope=None):
     """
@@ -26,7 +33,7 @@ def raw_rnn(cell, loop_fn, parallel_iterations=None, swap_memory=False, scope=No
         final cell state,
     )
     """
-    if not _like_rnncell(cell):
+    if not _is_rnn_cell(cell):
         raise TypeError("cell must be an instance of RNNCell")
     if not callable(loop_fn):
         raise TypeError("loop_fn must be a callable")
@@ -37,7 +44,12 @@ def raw_rnn(cell, loop_fn, parallel_iterations=None, swap_memory=False, scope=No
     # determined by the parent scope, or is set to place the cached
     # Variable using the same placement as for the rest of the RNN.
     with vs.variable_scope(scope or "rnn") as varscope:
-        if context.in_graph_mode():
+        in_graph_mode_fn = getattr(context, "in_graph_mode", None)
+        if in_graph_mode_fn is not None:
+            in_graph_mode = in_graph_mode_fn()
+        else:
+            in_graph_mode = not context.executing_eagerly()
+        if in_graph_mode:
             if varscope.caching_device is None:
                 varscope.set_caching_device(lambda op: op.device)
 
@@ -158,7 +170,7 @@ def copy_fn(cur_i, cand_i):
             return (next_time, elements_finished, next_input, state_ta,
                     emit_ta, next_state, loop_state)
 
-        returned = control_flow_ops.while_loop(
+        returned = tf.while_loop(
             condition, body, loop_vars=[
                 time, elements_finished, next_input, state_ta,
                 emit_ta, state, loop_state],
@@ -195,7 +207,7 @@ def loop_fn(time, cell_output, cell_state, loop_state):
         elements_finished = time >= sequence_length
         finished = math_ops.reduce_all(elements_finished)
 
-        next_input = control_flow_ops.cond(
+        next_input = tf.cond(
             finished,
             lambda: array_ops.zeros([array_ops.shape(inputs)[1], inputs.shape.as_list()[2]], dtype=dtypes.float32),
             lambda: inputs_ta.read(time)
@@ -233,7 +245,7 @@ def loop_fn(time, cell_output, cell_state, loop_state):
         )
         finished = math_ops.reduce_all(elements_finished)
 
-        next_input = control_flow_ops.cond(
+        next_input = tf.cond(
             finished,
             lambda: array_ops.zeros_like(initial_input),
             lambda: initial_input if cell_output is None else cell.output_function(next_cell_state)

tf_base_model.py

@@ -7,11 +7,14 @@
 import time
 
 import numpy as np
-import tensorflow as tf
+import tensorflow.compat.v1 as tf
 
 from tf_utils import shape
 
 
+tf.disable_v2_behavior()
+
+
 class TFBaseModel(object):
 
     """Interface containing some boilerplate code for training tensorflow models.

tf_utils.py

@@ -1,4 +1,8 @@
-import tensorflow as tf
+import tensorflow.compat.v1 as tf
+
+tf.disable_v2_behavior()
+
+from tensorflow.keras import initializers
 
 
 def dense_layer(inputs, output_units, bias=True, activation=None, batch_norm=None,
@@ -17,7 +21,7 @@ def dense_layer(inputs, output_units, bias=True, activation=None, batch_norm=Non
     with tf.variable_scope(scope, reuse=reuse):
         W = tf.get_variable(
             name='weights',
-            initializer=tf.contrib.layers.variance_scaling_initializer(),
+            initializer=initializers.VarianceScaling(),
             shape=[shape(inputs, -1), output_units]
         )
         z = tf.matmul(inputs, W)
@@ -33,7 +37,7 @@ def dense_layer(inputs, output_units, bias=True, activation=None, batch_norm=Non
             z = tf.layers.batch_normalization(z, training=batch_norm, reuse=reuse)
 
         z = activation(z) if activation else z
-        z = tf.nn.dropout(z, dropout) if dropout is not None else z
+        z = tf.nn.dropout(z, keep_prob=dropout) if dropout is not None else z
         return z
 
 
@@ -57,7 +61,7 @@ def time_distributed_dense_layer(
     with tf.variable_scope(scope, reuse=reuse):
         W = tf.get_variable(
             name='weights',
-            initializer=tf.contrib.layers.variance_scaling_initializer(),
+            initializer=initializers.VarianceScaling(),
             shape=[shape(inputs, -1), output_units]
         )
         z = tf.einsum('ijk,kl->ijl', inputs, W)
@@ -73,7 +77,7 @@ def time_distributed_dense_layer(
             z = tf.layers.batch_normalization(z, training=batch_norm, reuse=reuse)
 
         z = activation(z) if activation else z
-        z = tf.nn.dropout(z, dropout) if dropout is not None else z
+        z = tf.nn.dropout(z, keep_prob=dropout) if dropout is not None else z
         return z