feat(map-generator): use biome models for generating the biome layer

biomes/draw.py

@@ -2,26 +2,33 @@ import fire
 import matplotlib.pyplot as plt
 from matplotlib.collections import PatchCollection
 from matplotlib.patches import Circle, Patch
-from utils import logger
+from utils import logger, to_range
 from constants import BIOMES
 
 import pandas as pd
 import cartopy.crs as ccrs
 
-def draw(df, path=None):
+def draw(df, earth=True, width=23.22, height=13, only_draw=False, path=None):
     logger.debug('draw(df, %s)', path)
     biomes = {}
     biome_numbers = df['biome_num'].unique()
 
     for i, row in df.iterrows():
+        if earth:
             p = (row.longitude, row.latitude)
+        else:
+            p = (to_range(-180, 180, 0, width)(row.longitude), to_range(-90, 90, 0, height)(row.latitude))
+
         if row.biome_num in biomes:
             biomes[row.biome_num].append(p)
         else:
             biomes[row.biome_num] = [p]
 
+    if earth:
         ax = plt.axes(projection=ccrs.PlateCarree())
         ax.stock_img()
+    else:
+        ax = plt.gca()
 
     legend_handles = []
     for n in biome_numbers:
@@ -37,8 +44,11 @@ def draw(df, path=None):
 
     ax.autoscale_view()
     figure = plt.gcf()
-    figure.set_size_inches(23.22, 13)
+    figure.set_size_inches(width, height)
     figure.subplots_adjust(left=0.02, right=0.79)
+
+    if only_draw: return
+
     if path:
         plt.savefig(path)
     else:

biomes/map_generator.py

@@ -9,10 +9,15 @@ from io import BytesIO
 import pandas as pd
 from shapely.geometry import Point, MultiPoint
 from descartes import PolygonPatch
+from constants import INPUTS, SEASONS
+from draw import draw
+from train import A_params
+from model import Model
+from utils import *
 
 parameters = {
     'width': {
-        'default': 900,
+        'default': 700,
         'type': 'int',
     },
     'height': {
@@ -34,16 +39,22 @@ parameters = {
         'step': 0.01
     },
     'max_elevation': {
-        'default': 30,
+        'default': 1e4,
         'type': 'int',
         'min': 0,
-        'max': 50,
+        'max': 1e4,
+    },
+    'min_elevation': {
+        'default': -400,
+        'type': 'int',
+        'min': -1000,
+        'max': 0
     },
     'ground_noise': {
-        'default': 15,
+        'default': 1.1e4,
         'type': 'int',
         'min': 0,
-        'max': 50,
+        'max': 1e5,
     },
     'water_proportion': {
         'default': 0.6,
@@ -109,6 +120,16 @@ parameters = {
         'default': False,
         'type': 'bool'
     },
+    'mean_temperature': {
+        'default': -4.2,
+        'type': 'float',
+        'step': 1,
+    },
+    'mean_precipitation': {
+        'default': 45.24,
+        'type': 'float',
+        'step': 1,
+    },
     'seed': {
         'default': '',
         'type': 'int',
@@ -156,6 +177,9 @@ def bound_check(ground, point):
     elif y >= h:
         y = y - h
 
+    if x < 0 or x >= w or y < 0 or y >= h:
+        return bound_check(ground, (x, y))
+
     return (x, y)
 
 
@@ -190,7 +214,7 @@ def continent_agent(ground, position, size):
         if not is_ground(ground[x, y]) and in_range((x, y), position, size**2 * np.pi):
             trials = 0
             size -= 1
-            ground[x, y] = np.random.randint(1, p['ground_noise'])
+            ground[x, y] = np.random.randint(p['water_level'] + 1, p['ground_noise'])
         else:
             trials += 1
 
@@ -218,6 +242,7 @@ def random_elevate_agent(ground, position, height, size=p['mountain_area_elevati
 
 
 def mountain_agent(ground, position):
+    print('mountain_agent')
     if not away_from_sea(ground, position):
         return
 
@@ -283,11 +308,11 @@ def generate_map(biomes=False, **kwargs):
 
     ground = ndimage.gaussian_filter(ground, sigma=(1 - p['sharpness']) * 20)
 
-    for i in range(int(ground_size * p['mountain_ratio'] / p['max_elevation']**2)):
+    for i in range(int(ground_size * p['mountain_ratio'] / (p['max_elevation'] / 2))):
         position = (np.random.randint(0, width), np.random.randint(0, height))
         mountain_agent(ground, position)
 
-    norm = colors.Normalize(vmin=1)
+    norm = colors.Normalize(vmin=p['water_level'] + 1)
     greys = cm.get_cmap('Greys')
     greys.set_under(color=SEA_COLOR)
 
@@ -311,12 +336,6 @@ def generate_map(biomes=False, **kwargs):
 
     return figfile
 
-def to_range(omin, omax, nmin, nmax):
-    orange = omax - omin
-    nrange = nmax - nmin
-
-    return lambda x: ((x - omin) * nrange / orange) + nmin
-        
 
 def generate_biomes(ground):
     width, height = p['width'], p['height']
@@ -325,7 +344,6 @@ def generate_biomes(ground):
     width_to_longitude = to_range(0, width, -180, 180)
 
     print('generate_biomes')
-    INPUTS = ['elevation', 'distance_to_water', 'latitude']
     
     data = {}
     for col in ['longitude', 'latitude', 'elevation', 'distance_to_water']:
@@ -341,17 +359,24 @@ def generate_biomes(ground):
             data['latitude'].append(height_to_latitude(y))
             data['elevation'].append(v)
 
+    print(len(points))
+    print('buffering points')
     points = MultiPoint(points)
-    boundary = points.buffer(1e-0).boundary
+    boundary = points.buffer(1).boundary
 
     for x, y in np.ndindex(ground.shape):
         if ground[x,y] > p['water_level']:
-            # print(x,y, Point(x,y).distance(boundary))
             data['distance_to_water'].append(Point(x, y).distance(boundary))
 
     
     df = pd.DataFrame(data)
-    print(df)
+    print(df['elevation'].min(), df['elevation'].max())
+    print(df['distance_to_water'].min(), df['distance_to_water'].max())
+    print(df['latitude'].min(), df['latitude'].max())
+
+    print('running prediction models')
+    print(p['mean_precipitation'], p['mean_temperature'])
+    result = predict_end_to_end(df, boundary)
 
     # fig = plt.figure()
     # ax = fig.add_subplot(111)
@@ -365,7 +390,118 @@ def generate_biomes(ground):
 
     # plt.show()
 
+df = pd.read_pickle('data.p')
+print(df['elevation'].min(), df['elevation'].max())
+print(df['distance_to_water'].min(), df['distance_to_water'].max())
+print(df['latitude'].min(), df['latitude'].max())
 
+def predict_end_to_end(input_df, boundary, checkpoint_temp='checkpoints/temp.h5', checkpoint_precip='checkpoints/precip.h5', checkpoint_biomes='checkpoints/b.h5', year=2000):
+    batch_size = A_params['batch_size']['grid_search'][0]
+    layers = A_params['layers']['grid_search'][0]
+    optimizer = A_params['optimizer']['grid_search'][0](A_params['lr']['grid_search'][0])
+
+    Temp = Model('temp', epochs=1)
+    Temp.prepare_for_use(
+        batch_size=batch_size,
+        layers=layers,
+        dataset_fn=dataframe_to_dataset_temp,
+        optimizer=optimizer,
+        out_activation=None,
+        loss='mse',
+        metrics=['mae']
+    )
+    Temp.restore(checkpoint_temp)
+
+    Precip = Model('precip', epochs=1)
+    Precip.prepare_for_use(
+        batch_size=batch_size,
+        layers=layers,
+        dataset_fn=dataframe_to_dataset_temp,
+        optimizer=optimizer,
+        out_activation=None,
+        loss='mse',
+        metrics=['mae']
+    )
+    Precip.restore(checkpoint_precip)
+
+    Biomes = Model('b', epochs=1)
+    Biomes.prepare_for_use()
+    Biomes.restore(checkpoint_biomes)
+
+    inputs = input_df[INPUTS]
+
+    inputs.loc[:, 'mean_temp'] = p['mean_temperature']
+    inputs_copy = inputs.copy()
+    inputs_copy.loc[:, 'mean_temp'] = mean_temperature_over_years(df, size=inputs.shape[0])
+
+    inputs = inputs.to_numpy()
+    inputs = normalize_ndarray(inputs, inputs_copy)
+    print(inputs)
+    out_columns = ['temp_{}_{}'.format(season, year) for season in SEASONS]
+    out = Temp.predict(inputs)
+    temp_output = pd.DataFrame(data=denormalize(out, df[out_columns].to_numpy()), columns=out_columns)
+
+    inputs = input_df[INPUTS]
+
+    inputs.loc[:, 'mean_precip'] = p['mean_precipitation']
+    inputs_copy = inputs.copy()
+    inputs_copy.loc[:, 'mean_precip'] = mean_precipitation_over_years(df, size=inputs.shape[0])
+
+    inputs = inputs.to_numpy()
+    inputs = normalize_ndarray(inputs, inputs_copy)
+    print(inputs)
+    out_columns = ['precip_{}_{}'.format(season, year) for season in SEASONS]
+    out = Precip.predict(inputs)
+
+    precip_output = pd.DataFrame(data=denormalize(out, df[out_columns].to_numpy()), columns=out_columns)
+
+    inputs = list(INPUTS)
+
+    frame = input_df[inputs + ['longitude']]
+
+    for season in SEASONS:
+        tc = 'temp_{}_{}'.format(season, year)
+        pc = 'precip_{}_{}'.format(season, year)
+        frame.loc[:, tc] = temp_output[tc]
+        frame.loc[:, pc] = precip_output[pc]
+
+    frame.loc[:, 'latitude'] = input_df['latitude']
+
+    frame_cp = frame.copy()
+
+    columns = ['latitude', 'longitude', 'biome_num']
+    new_data = pd.DataFrame(columns=columns)
+    nframe = pd.DataFrame(columns=frame.columns, data=normalize_ndarray(frame.to_numpy()))
+
+    for season in SEASONS:
+        inputs += [
+            'temp_{}_{}'.format(season, year),
+            'precip_{}_{}'.format(season, year)
+        ]
+
+    for i, (chunk, chunk_original) in enumerate(zip(chunker(nframe, Biomes.batch_size), chunker(frame_cp, Biomes.batch_size))):
+        if chunk.shape[0] < Biomes.batch_size:
+            continue
+        input_data = chunk.loc[:, inputs].values
+        out = Biomes.predict_class(input_data)
+
+        f = pd.DataFrame({
+            'longitude': chunk_original.loc[:, 'longitude'], 
+            'latitude': chunk_original.loc[:, 'latitude'],
+            'biome_num': out
+        }, columns=columns)
+        new_data = new_data.append(f)
+
+    #print(new_data)
+    draw(new_data, earth=False, only_draw=True, width=p['width'], height=p['height'])
+
+# TODO: reduce opacity of biome layer
 if __name__ == "__main__":
-    generate_map()
+    # p['width'] = 50
+    # p['height'] = 50
+    p['water_proportion'] = 0.9
+    p['continents'] = 3
+    p['seed'] = 1
+    generate_map(True)
+    # print(normalize_ndarray(np.array([[ 5.59359803,0.99879546,-90., 45.24], [  5.54976747, 0.99879546,-86.4, 45.24      ]])))
     plt.show()

biomes/predict.py

@@ -147,6 +147,7 @@ def predict_end_to_end(Temp, Precip, Biomes, year=2000):
 
     all_temps = ['temp_{}_{}'.format(season, year) for season in SEASONS]
     inputs.loc[:, 'mean_temp'] = np.mean(df[all_temps].values)
+    print(inputs['mean_temp'])
 
     inputs = inputs.to_numpy()
     inputs = normalize_ndarray(inputs)
@@ -158,6 +159,7 @@ def predict_end_to_end(Temp, Precip, Biomes, year=2000):
 
     all_precips = ['precip_{}_{}'.format(season, year) for season in SEASONS]
     inputs.loc[:, 'mean_precip'] = np.mean(df[all_precips].values)
+    print(inputs['mean_precip'])
 
     inputs = inputs.to_numpy()
     inputs = normalize_ndarray(inputs)
@@ -168,12 +170,6 @@ def predict_end_to_end(Temp, Precip, Biomes, year=2000):
 
     inputs = list(INPUTS)
 
-    for season in SEASONS:
-        inputs += [
-            'temp_{}_{}'.format(season, year),
-            'precip_{}_{}'.format(season, year)
-        ]
-
     frame = df[inputs + ['longitude']]
 
     for season in SEASONS:
@@ -190,6 +186,12 @@ def predict_end_to_end(Temp, Precip, Biomes, year=2000):
     new_data = pd.DataFrame(columns=columns)
     nframe = pd.DataFrame(columns=frame.columns, data=normalize_ndarray(frame.to_numpy()))
 
+    for season in SEASONS:
+        inputs += [
+            'temp_{}_{}'.format(season, year),
+            'precip_{}_{}'.format(season, year)
+        ]
+
     for i, (chunk, chunk_original) in enumerate(zip(chunker(nframe, Biomes.batch_size), chunker(frame_cp, Biomes.batch_size))):
         if chunk.shape[0] < Biomes.batch_size:
             continue

biomes/utils.py

@@ -6,15 +6,17 @@ from sklearn.utils import class_weight
 from constants import *
 import logging
 import os
+from math import ceil
 
 logger = logging.getLogger('main')
 logger.setLevel(os.environ.get('LOG_LEVEL', 'INFO'))
 
+EPSILON = 1e-5
 
 def normalize(v, o=None):
     if o is None:
         o = v
-    return (v - np.mean(o)) / np.std(o)
+    return (v - np.mean(o)) / max(EPSILON, np.std(o))
 
 def denormalize(v, o=None):
     if o is None:
@@ -132,8 +134,28 @@ def dataframe_to_dataset_precip(df):
     logger.debug('dataset size: rows=%d, input_columns=%d, num_classes=%d', int(tf_inputs.shape[0]), input_columns, num_classes)
     return int(tf_inputs.shape[0]), input_columns, num_classes, None, tf.data.Dataset.from_tensor_slices((tf_inputs, tf_output))
 
+def mean_temperature_over_years(df, size=MAX_YEAR - MIN_YEAR):
+    means = []
+    for year in range(MIN_YEAR, MAX_YEAR + 1):
+        all_temps = ['temp_{}_{}'.format(season, year) for season in SEASONS]
+        means.append(np.mean(df[all_temps].values))
+    return (means * ceil(size / len(means)))[0:size]
+
+def mean_precipitation_over_years(df, size=MAX_YEAR - MIN_YEAR):
+    means = []
+    for year in range(MIN_YEAR, MAX_YEAR + 1):
+        all_precips = ['precip_{}_{}'.format(season, year) for season in SEASONS]
+        means.append(np.mean(df[all_precips].values))
+    return (means * ceil(size / len(means)))[0:size]
 
 flatten = lambda l: [item for sublist in l for item in sublist]
 
 def chunker(seq, size):
     return (seq[pos:pos + size] for pos in range(0, len(seq), size))
+
+def to_range(omin, omax, nmin, nmax):
+    orange = omax - omin
+    nrange = nmax - nmin
+
+    return lambda x: ((x - omin) * nrange / orange) + nmin
+        

biomes/web.py

@@ -1,5 +1,5 @@
 from flask import Flask, render_template, make_response, send_file, request
-from index import generate_map, parameters
+from map_generator import generate_map, parameters
 
 app = Flask(__name__)